Skip to content
Permalink
master
Switch branches/tags
Go to file
 
 
Cannot retrieve contributors at this time
10123 lines (9376 sloc) 613 KB

Release 2.10.0

Breaking Changes

  • <THIS SECTION SHOULD CONTAIN API, ABI AND BEHAVIORAL BREAKING CHANGES>
  • Some files in tensorflow/python/training have been moved to tensorflow/python/tracking and tensorflow/python/checkpoint. Please update your imports accordingly, the old files will be removed in Release 2.11.
  • RNG behavior change for tf.keras.initializers. Keras initializers will now use stateless random ops to generate random numbers.
    • Both seeded and unseeded initializers will always generate the same values every time they are called (for a given variable shape). For unseeded initializers (seed=None), a random seed will be created and assigned at initializer creation (different initializer instances get different seeds).
    • An unseeded initializer will raise a warning if it is reused (called) multiple times. This is because it would produce the same values each time, which may not be intended.

Known Caveats

  • <CAVEATS REGARDING THE RELEASE (BUT NOT BREAKING CHANGES).>
  • <ADDING/BUMPING DEPENDENCIES SHOULD GO HERE>
  • <KNOWN LACK OF SUPPORT ON SOME PLATFORM, SHOULD GO HERE>

Major Features and Improvements

  • tf.lite:

    • New operations supported:
      • tflite SelectV2 now supports 5D.
      • tf.einsum is supported with multiple unknown shapes.
      • tf.unsortedsegmentprod op is supported.
    • Upgrade Flatbuffers v2.0.5 from v1.12.0
  • tf.keras:

    • Added tf.keras.models.experimental.SharpnessAwareMinimization. This class implements the sharpness-aware minimization technique, which boosts model performance on various tasks, e.g., ResNet on image classification.
    • EinsumDense layer moved from experimental to core. Its import path moved from tf.keras.layers.experimental.EinsumDense to tf.keras.layers.EinsumDense.
    • Added tf.keras.utils.audio_dataset_from_directory utility to easily generate audio classification datasets from directories of .wav files.
    • Added subset="both" support in tf.keras.utils.image_dataset_from_directory, tf.keras.utils.text_dataset_from_directory, and audio_dataset_from_directory, to be used with the validation_split argument, for returning both dataset splits at once, as a tuple.
    • Added tf.keras.utils.split_dataset utility to split a Dataset object or a list/tuple of arrays into two Dataset objects (e.g. train/test).
    • Added step granularity to BackupAndRestore callback for handling distributed training failures & restarts. The training state can now be restored at the exact epoch and step at which it was previously saved before failing.
    • Added tf.keras.dtensor.experimental.optimizers.AdamW. This optimizer is similar as the existing keras.optimizers.experimental.AdamW, and works in the DTensor training use case.
  • tf.data:

    • Added support for cross-trainer data caching in tf.data service. This saves computation resources when concurrent training jobs train from the same dataset. See https://www.tensorflow.org/api_docs/python/tf/data/experimental/service#sharing_tfdata_service_with_concurrent_trainers for more details.
    • Added a new field, inject_prefetch, to tf.data.experimental.OptimizationOptions. If it is set to True, tf.data will now automatically add a prefetch transformation to datasets that end in synchronous transformations. This enables data generation to be overlapped with data consumption. This may cause a small increase in memory usage due to buffering. To enable this behavior, set inject_prefetch=True in tf.data.experimental.OptimizationOptions.
    • Added a new value to tf.data.Options.autotune.autotune_algorithm: STAGE_BASED. If the autotune algorithm is set to STAGE_BASED, then it runs a new algorithm that can get the same performance with lower CPU/memory usage.
  • tf.distribute:

    • Added tf.distribute.experimental.PreemptionCheckpointHandler to handle worker preemption/maintenance and cluster-wise consistent error reporting for tf.distribute.MultiWorkerMirroredStrategy. Specifically, for the type of interruption with advance notice, it automatically saves a checkpoint, exits the program without raising an unrecoverable error, and restores the progress when training restarts.
  • tf.math:

    • Added tf.math.approx_max_k and tf.math.approx_min_k which are the optimized alternatives to tf.math.top_k on TPU. The performance difference range from 8 to 100 times depending on the size of k. When running on CPU and GPU, a non-optimized XLA kernel is used.
  • tf.vectorized_map:

    • Added an optional parameter: warn. This parameter controls whether or not warnings will be printed when operations in the provided fn fall back to a while loop.
  • XLA:

    • MWMS is now compilable with XLA.

Bug Fixes and Other Changes

  • New argument experimental_device_ordinal in LogicalDeviceConfiguration to control the order of logical devices. (GPU only)

  • tf.keras:

    • Changed the TensorBoard tag names produced by the tf.keras.callbacks.TensorBoard callback, so that summaries logged automatically for model weights now include either a /histogram or /image suffix in their tag names, in order to prevent tag name collisions across summary types.
  • When running on GPU (with cuDNN version 7.6.3 or later), tf.nn.depthwise_conv2d backprop to filter (and therefore also tf.keras.layers.DepthwiseConv2D) now operate deterministically (and tf.errors.UnimplementedError is no longer thrown) when op-determinism has been enabled via tf.config.experimental.enable_op_determinism. This closes issue 47174.

  • tf.random

    • Added tf.random.experimental.stateless_shuffle, a stateless version of tf.random.shuffle.

Deprecations

  • The C++ tensorflow::Code and tensorflow::Status will become aliases of respectively absl::StatusCode and absl::Status in some future release.
    • Use tensorflow::OkStatus() instead of tensorflow::Status::OK().
    • Stop constructing Status objects from tensorflow::error::Code.
    • One MUST NOT access tensorflow::errors::Code fields. Accessing tensorflow::error::Code fields is fine.
      • Use the constructors such as tensorflow::errors:InvalidArgument to create status using an error code without accessing it.
      • Use the free functions such as tensorflow::errors::IsInvalidArgument if needed.
      • In the last resort, use e.g. static_cast<tensorflow::errors::Code>(error::Code::INVALID_ARGUMENT) or static_cast<int>(code) for comparisons.
  • tensorflow::StatusOr will also become in the future alias to absl::StatusOr, so use StatusOr::value instead of StatusOr::ConsumeValueOrDie.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

, , , , ,

Release 2.9.1

Add an upper bound for protobuf in setup.py since protobuf after version 3.20 is currently incompatible with TensorFlow. See #53234, protocolbuffers/protobuf#9954 and #56077.

Release 2.8.2

Add an upper bound for protobuf in setup.py since protobuf after version 3.20 is currently incompatible with TensorFlow. See #53234, protocolbuffers/protobuf#9954 and #56077.

Release 2.7.3

Add an upper bound for protobuf in setup.py since protobuf after version 3.20 is currently incompatible with TensorFlow. See #53234, protocolbuffers/protobuf#9954 and #56077.

Release 2.6.5

Add an upper bound for protobuf in setup.py since protobuf after version 3.20 is currently incompatible with TensorFlow. See #53234, protocolbuffers/protobuf#9954 and #56077.

Release 2.9.0

Breaking Changes

  • Due to security issues in TF 2.8, all boosted trees code has now been removed (after being deprecated in TF 2.8). Users should switch to TensorFlow Decision Forests.
  • Build, Compilation and Packaging
    • TensorFlow is now compiled with _GLIBCXX_USE_CXX11_ABI=1. Downstream projects that encounter std::__cxx11 or [abi:cxx11] linker errors will need to adopt this compiler option. See the GNU C++ Library docs on Dual ABI.
    • TensorFlow Python wheels now specifically conform to manylinux2014, an upgrade from manylinux2010. The minimum Pip version supporting manylinux2014 is Pip 19.3 (see pypa/manylinux. This change may affect you if you have been using TensorFlow on a very old platform equivalent to CentOS 6, as manylinux2014 targets CentOS 7 as a compatibility base. Note that TensorFlow does not officially support either platform.
    • Discussion for these changes can be found on SIG Build's TensorFlow Community Forum thread
  • The tf.keras.mixed_precision.experimental API has been removed. The non-experimental symbols under tf.keras.mixed_precision have been available since TensorFlow 2.4 and should be used instead.
    • The non-experimental API has some minor differences from the experimental API. In most cases, you only need to make three minor changes:
      • Remove the word "experimental" from tf.keras.mixed_precision symbols. E.g., replace tf.keras.mixed_precision.experimental.global_policy with tf.keras.mixed_precision.global_policy.
      • Replace tf.keras.mixed_precision.experimental.set_policy with tf.keras.mixed_precision.set_global_policy. The experimental symbol set_policy was renamed to set_global_policy in the non-experimental API.
      • Replace LossScaleOptimizer(opt, "dynamic") with LossScaleOptimizer(opt). If you pass anything other than "dynamic" to the second argument, see (1) of the next section.
    • In the following rare cases, you need to make more changes when switching to the non-experimental API:
      • If you passed anything other than "dynamic" to the loss_scale argument (the second argument) of LossScaleOptimizer:
      • If you passed a value to the loss_scale argument (the second argument) of Policy:
        • The experimental version of Policy optionally took in a tf.compat.v1.mixed_precision.LossScale in the constructor, which defaulted to a dynamic loss scale for the "mixed_float16" policy and no loss scale for other policies. In Model.compile, if the model's policy had a loss scale, the optimizer would be wrapped with a LossScaleOptimizer. With the non-experimental Policy, there is no loss scale associated with the Policy, and Model.compile wraps the optimizer with a LossScaleOptimizer if and only if the policy is a "mixed_float16" policy. If you previously passed a LossScale to the experimental Policy, consider just removing it, as the default loss scaling behavior is usually what you want. If you really want to customize the loss scaling behavior, you can wrap your optimizer with a LossScaleOptimizer before passing it to Model.compile.
      • If you use the very rarely-used function tf.keras.mixed_precision.experimental.get_layer_policy:
        • Replace tf.keras.mixed_precision.experimental.get_layer_policy(layer) with layer.dtype_policy.
  • tf.mixed_precision.experimental.LossScale and its subclasses have been removed from the TF2 namespace. This symbols were very rarely used and were only useful in TF2 for use in the now-removed tf.keras.mixed_precision.experimental API. The symbols are still available under tf.compat.v1.mixed_precision.
  • The experimental_relax_shapes heuristic for tf.function has been deprecated and replaced with reduce_retracing which encompasses broader heuristics to reduce the number of retraces (see below)

Major Features and Improvements

  • tf.keras:

    • Added tf.keras.applications.resnet_rs models. This includes the ResNetRS50, ResNetRS101, ResNetRS152, ResNetRS200, ResNetRS270, ResNetRS350 and ResNetRS420 model architectures. The ResNetRS models are based on the architecture described in Revisiting ResNets: Improved Training and Scaling Strategies
    • Added tf.keras.optimizers.experimental.Optimizer. The reworked optimizer gives more control over different phases of optimizer calls, and is easier to customize. We provide Adam, SGD, Adadelta, AdaGrad and RMSprop optimizers based on tf.keras.optimizers.experimental.Optimizer. Generally the new optimizers work in the same way as the old ones, but support new constructor arguments. In the future, the symbols tf.keras.optimizers.Optimizer/Adam/etc will point to the new optimizers, and the previous generation of optimizers will be moved to tf.keras.optimizers.legacy.Optimizer/Adam/etc.
    • Added L2 unit normalization layer tf.keras.layers.UnitNormalization.
    • Added tf.keras.regularizers.OrthogonalRegularizer, a new regularizer that encourages orthogonality between the rows (or columns) or a weight matrix.
    • Added tf.keras.layers.RandomBrightness layer for image preprocessing.
    • Added APIs for switching between interactive logging and absl logging. By default, Keras always writes the logs to stdout. However, this is not optimal in a non-interactive environment, where you don't have access to stdout, but can only view the logs. You can use tf.keras.utils.disable_interactive_logging() to write the logs to ABSL logging. You can also use tf.keras.utils.enable_interactive_logging() to change it back to stdout, or tf.keras.utils.is_interactive_logging_enabled() to check if interactive logging is enabled.
    • Changed default value for the verbose argument of Model.evaluate() and Model.predict() to "auto", which defaults to verbose=1 for most cases and defaults to verbose=2 when used with ParameterServerStrategy or with interactive logging disabled.
    • Argument jit_compile in Model.compile() now applies to Model.evaluate() and Model.predict(). Setting jit_compile=True in compile() compiles the model's training, evaluation, and inference steps to XLA. Note that jit_compile=True may not necessarily work for all models.
    • Added DTensor-related Keras APIs under tf.keras.dtensor namespace. The APIs are still classified as experimental. You are welcome to try it out. Please check the tutorial and guide on https://www.tensorflow.org/ for more details about DTensor.
  • tf.lite:

    • Added TFLite builtin op support for the following TF ops:
      • tf.math.argmin/tf.math.argmax for input data type tf.bool on CPU.
      • tf.nn.gelu op for output data type tf.float32 and quantization on CPU.
    • Add nominal support for unsigned 16-bit integer tensor types. Note that very few TFLite kernels support this type natively, so its use in mobile ML authoring is generally discouraged.
    • Add support for unsigned 16-bit integer tensor types in cast op.
    • Experimental support for lowering list_ops.tensor_list_set_item with DynamicUpdateSlice.
    • Enabled a new MLIR-based dynamic range quantization backend by default
      • The new backend is used for post-training int8 dynamic range quantization and post-training float16 quantization.
      • Set experimental_new_dynamic_range_quantizer in tf.lite.TFLiteConverter to False to disable this change
    • Native TF Lite variables are now enabled during conversion by default on all v2 TfLiteConverter entry points. experimental_enable_resource_variables on tf.lite.TFLiteConverter is now True by default and will be removed in the future.
  • tf.function:

    • Custom classes used as arguments for tf.function can now specify rules regarding when retracing needs to occur by implementing the Tracing Protocol available through tf.types.experimental.SupportsTracingProtocol.
    • TypeSpec classes (as associated with ExtensionTypes) also implement the Tracing Protocol which can be overridden if necessary.
    • The newly introduced reduce_retracing option also uses the Tracing Protocol to proactively generate generalized traces similar to experimental_relax_shapes (which has now been deprecated).
  • Unified eager and tf.function execution:

    • Eager mode can now execute each op as a tf.function, allowing for more consistent feature support in future releases.
    • It is available for immediate use.
      • See the TF_RUN_EAGER_OP_AS_FUNCTION environment variable in eager context.
      • Eager performance should be similar with this feature enabled.
        • A roughly 5us per-op overhead may be observed when running many small functions.
        • Note a known issue with GPU performance.
      • The behavior of tf.function itself is unaffected.
    • Note: This feature will be enabled by default in an upcoming version of TensorFlow.
  • tf.experimental.dtensor: Added DTensor, an extension to TensorFlow for large-scale modeling with minimal changes to user code. You are welcome to try it out, though be aware that the DTensor API is experimental and up-to backward-incompatible changes. DTensor and Keras integration is published under tf.keras.dtensor in this release (refer to the tf.keras entry). The tutoral and guide for DTensor will be published on https://www.tensorflow.org/. Please stay tuned.

  • oneDNN CPU performance optimizations are available in Linux x86, Windows x86, and Linux aarch64 packages.

    • Linux x86 packages:
      • oneDNN optimizations are enabled by default on CPUs with neural-network-focused hardware features such as AVX512_VNNI, AVX512_BF16, AMX, etc. (Intel Cascade Lake and newer CPUs.)
      • For older CPUs, oneDNN optimizations are disabled by default.
    • Windows x86 package: oneDNN optimizations are disabled by default.
    • Linux aach64 (--config=mkl_aarch64) package:
      • Experimental oneDNN optimizations are disabled by default.
      • If you experience issues with oneDNN optimizations on, we recommend turning them off.
    • To explicitly enable or disable oneDNN optimizations, set the environment variable TF_ENABLE_ONEDNN_OPTS to 1 (enable) or 0 (disable) before running TensorFlow. (The variable is checked during import tensorflow.) To fall back to default settings, unset the environment variable.
    • These optimizations can yield slightly different numerical results from when they are off due to floating-point round-off errors from different computation approaches and orders.
    • To verify that the optimizations are on, look for a message with "oneDNN custom operations are on" in the log. If the exact phrase is not there, it means they are off.

Bug Fixes and Other Changes

  • tf.data:

    • Fixed bug in tf.data.experimental.parse_example_dataset when tf.io.RaggedFeatures would specify value_key but no partitions. Before the fix, setting value_key but no partitions would result in the feature key being replaced by the value key, e.g. {'value_key': <RaggedTensor>} instead of {'key': <RaggedTensor>}. Now the correct feature key will be used. This aligns the behavior of tf.data.experimental.parse_example_dataset to match the behavior of tf.io.parse_example.
    • Added a new field, filter_parallelization, to tf.data.experimental.OptimizationOptions. If it is set to True, tf.data will run Filter transformation with multiple threads. Its default value is False if not specified.
  • tf.keras:

    • Fixed bug in optimizers that prevented them from properly checkpointing slot variables when they are ShardedVariables (used for training with tf.distribute.experimental.ParameterServerStrategy).
  • tf.random:

    • Added tf.random.experimental.index_shuffle, for shuffling a sequence without materializing the sequence in memory.
  • tf.RaggedTensor:

    • Introduced tf.experimental.RowPartition, which encodes how one dimension in a RaggedTensor relates to another, into the public API.
    • Introduced tf.experimental.DynamicRaggedShape, which represents the shape of a RaggedTensor.

Security

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Aaron Debattista, Abel Soares Siqueira, Abhishek Varma, Andrei Ivanov, andreii, Andrew Goodbody, apeltop, Arnab Dutta, Ashiq Imran, Banikumar Maiti (Intel Aipg), Ben Greiner, Benjamin Peterson, bhack, Christopher Bate, chunduriv, Copybara-Service, DEKHTIARJonathan, Deven Desai, Duncan Riach, Eric Kunze, Everton Constantino, Faruk D, Fredrik Knutsson, gadagashwini, Gauri1 Deshpande, gtiHibGele, Guozhong Zhuang, Islem-Esi, Ivanov Viktor, Jason Furmanek, Jason Zaman, Jim, Jinzhe Zeng, John Laxson, Jonas Eschle, Jonas Eschle 'Mayou36, Jonathan Dekhtiar, Kaixi Hou, Kanvi Khanna, KaurkerDevourer, Koan-Sin Tan, kushanam, Laramie Leavitt, Li-Wen Chang, lipracer, Louis Sugy, Lu Teng, Mahmoud Abuzaina, Malcolm Slaney, Malik Shahzad Muzaffar, Marek Šuppa, Matt Conley, Michael Melesse, Milos Puzovic, mohantym, Nathan John Sircombe, Nathan Luehr, Nilesh Agarwalla, Patrice Vignola, peterjc123, Philip Turner, Rajeshwar Reddy T, Robert Kalmar, Rodrigo Formigone, Rohit Santhanam, rui, Sachin Muradi, Saduf2019, sandip, Scott Leishman, Serge Panev, Shi,Guangyong, Srinivasan Narayanamoorthy, stanley, Steven I Reeves, stevenireeves, sushreebarsa, Tamas Bela Feher, Tao He, Thomas Schmeyer, Tiago Almeida, Trevor Morris, Uday Bondhugula, Uwe L. Korn, Varghese, Jojimon, Vishnuvardhan Janapati, William Muir, William Raveane, xutianming, Yasuhiro Matsumoto, Yimei Sun, Yong Tang, Yu Feng, Yuriy Chernyshov, zhaozheng09

Release 2.8.1

This releases introduces several vulnerability fixes:

Release 2.7.2

This releases introduces several vulnerability fixes:

Release 2.6.4

This releases introduces several vulnerability fixes:

Release 2.8.0

Major Features and Improvements

  • tf.lite:

    • Added TFLite builtin op support for the following TF ops:
      • tf.raw_ops.Bucketize op on CPU.
      • tf.where op for data types tf.int32/tf.uint32/tf.int8/tf.uint8/tf.int64.
      • tf.random.normal op for output data type tf.float32 on CPU.
      • tf.random.uniform op for output data type tf.float32 on CPU.
      • tf.random.categorical op for output data type tf.int64 on CPU.
  • tensorflow.experimental.tensorrt:

    • conversion_params is now deprecated inside TrtGraphConverterV2 in favor of direct arguments: max_workspace_size_bytes, precision_mode, minimum_segment_size, maximum_cached_engines, use_calibration and allow_build_at_runtime.
    • Added a new parameter called save_gpu_specific_engines to the .save() function inside TrtGraphConverterV2. When False, the .save() function won't save any TRT engines that have been built. When True (default), the original behavior is preserved.
    • TrtGraphConverterV2 provides a new API called .summary() which outputs a summary of the inference converted by TF-TRT. It namely shows each TRTEngineOp with their input(s)' and output(s)' shape and dtype. A detailed version of the summary is available which prints additionally all the TensorFlow OPs included in each of the TRTEngineOps.
  • tf.tpu.experimental.embedding:

    • tf.tpu.experimental.embedding.FeatureConfig now takes an additional argument output_shape which can specify the shape of the output activation for the feature.
    • tf.tpu.experimental.embedding.TPUEmbedding now has the same behavior as tf.tpu.experimental.embedding.serving_embedding_lookup which can take arbitrary rank of dense and sparse tensor. For ragged tensor, though the input tensor remains to be rank 2, the activations now can be rank 2 or above by specifying the output shape in the feature config or via the build method.
  • Add tf.config.experimental.enable_op_determinism, which makes TensorFlow ops run deterministically at the cost of performance. Replaces the TF_DETERMINISTIC_OPS environmental variable, which is now deprecated. The "Bug Fixes and Other Changes" section lists more determinism-related changes.

  • (Since TF 2.7) Add PluggableDevice support to TensorFlow Profiler.

Bug Fixes and Other Changes

  • tf.data:

    • Fixed a bug where setting options.deterministic = False would only modify one transformation to run non-deterministically, leaving other transformations deterministic. The option will now apply the same across all transformations.
    • The optimization parallel_batch now becomes default if not disabled by users, which will parallelize copying of batch elements.
    • Added the ability for TensorSliceDataset to identify and handle inputs that are files. This enables creating hermetic SavedModels when using datasets created from files.
  • tf.lite:

    • Adds GPU Delegation support for serialization to Java API. This boosts initialization time up to 90% when OpenCL is available.
    • Deprecated Interpreter::SetNumThreads, in favor of InterpreterBuilder::SetNumThreads.
  • tf.keras:

    • Adds tf.compat.v1.keras.utils.get_or_create_layer to aid migration to TF2 by enabling tracking of nested keras models created in TF1-style, when used with the tf.compat.v1.keras.utils.track_tf1_style_variables decorator.
    • Added a tf.keras.layers.experimental.preprocessing.HashedCrossing layer which applies the hashing trick to the concatenation of crossed scalar inputs. This provides a stateless way to try adding feature crosses of integer or string data to a model.
    • Removed keras.layers.experimental.preprocessing.CategoryCrossing. Users should migrate to the HashedCrossing layer or use tf.sparse.cross/tf.ragged.cross directly.
    • Added additional standardize and split modes to TextVectorization:
      • standardize="lower" will lowercase inputs.
      • standardize="string_punctuation" will remove all punctuation.
      • split="character" will split on every unicode character.
    • Added an output_mode argument to the Discretization and Hashing layers with the same semantics as other preprocessing layers. All categorical preprocessing layers now support output_mode.
    • All preprocessing layer output will follow the compute dtype of a tf.keras.mixed_precision.Policy, unless constructed with output_mode="int" in which case output will be tf.int64. The output type of any preprocessing layer can be controlled individually by passing a dtype argument to the layer.
    • tf.random.Generator for keras initializers and all RNG code.
    • Added 3 new APIs for enable/disable/check the usage of tf.random.Generator in keras backend, which will be the new backend for all the RNG in Keras. We plan to switch on the new code path by default in tf 2.8, and the behavior change will likely to cause some breakage on user side (eg if the test is checking against some golden number). These 3 APIs will allow user to disable and switch back to legacy behavior if they prefer. In future (eg TF 2.10), we expect to totally remove the legacy code path (stateful random Ops), and these 3 APIs will be removed as well.
    • tf.keras.callbacks.experimental.BackupAndRestore is now available as tf.keras.callbacks.BackupAndRestore. The experimental endpoint is deprecated and will be removed in a future release.
    • tf.keras.experimental.SidecarEvaluator is now available as tf.keras.utils.SidecarEvaluator. The experimental endpoint is deprecated and will be removed in a future release.
    • Metrics update and collection logic in default Model.train_step() is now customizable via overriding Model.compute_metrics().
    • Losses computation logic in default Model.train_step() is now customizable via overriding Model.compute_loss().
    • jit_compile added to Model.compile() on an opt-in basis to compile the model's training step with XLA. Note that jit_compile=True may not necessarily work for all models.
  • Deterministic Op Functionality:

    • Fix regression in deterministic selection of deterministic cuDNN convolution algorithms, a regression that was introduced in v2.5. Note that nondeterministic out-of-memory events while selecting algorithms could still lead to nondeterminism, although this is very unlikely. This additional, unlikely source will be eliminated in a later version.
    • Add deterministic GPU implementations of:
      • tf.function(jit_compile=True)'s that use Scatter.
      • (since v2.7) Stateful ops used in tf.data.Dataset
      • (since v2.7) tf.convert_to_tensor when fed with (sparse) tf.IndexedSlices (because it uses tf.math.unsorted_segment_sum)
      • (since v2.7) tf.gather backprop (because tf.convert_to_tensor reduces tf.gather's (sparse) tf.IndexedSlices gradients into its dense params input)
      • (since v2.7) tf.math.segment_mean
      • (since v2.7) tf.math.segment_prod
      • (since v2.7) tf.math.segment_sum
      • (since v2.7) tf.math.unsorted_segment_mean
      • (since v2.7) tf.math.unsorted_segment_prod
      • (since v2.7) tf.math.unsorted_segment_sum
      • (since v2.7) tf.math.unsorted_segment_sqrt
      • (since v2.7) tf.nn.ctc_loss (resolved, possibly in prior release, and confirmed with tests)
      • (since v2.7)tf.nn.sparse_softmax_crossentropy_with_logits
    • (since v2.7) Run tf.scatter_nd and other related scatter functions, such as tf.tensor_scatter_nd_update, on CPU (with significant performance penalty).
    • Add determinism-unimplemented exception-throwing to the following ops. When op-determinism is expected (i.e. after tf.config.experimental.enable_op_determinism has been called), an attempt to use the specified paths through the following ops on a GPU will cause tf.errors.UnimplementedError (with an understandable message), unless otherwise specified, to be thrown.
      • FakeQuantWithMinMaxVarsGradient and FakeQuantWithMinMaxVarsPerChannelGradient
      • (since v2.7) tf.compat.v1.get_seed if the global random seed has not yet been set (via tf.random.set_seed). Throws RuntimeError from Python or InvalidArgument from C++
      • (since v2.7) tf.compat.v1.nn.fused_batch_norm backprop to offset when is_training=False
      • (since v2.7) tf.image.adjust_contrast forward
      • (since v2.7) tf.image.resize with method=ResizeMethod.NEAREST backprop
      • (since v2.7) tf.linalg.svd
      • (since v2.7) tf.math.bincount
      • (since v2.7) tf.nn.depthwise_conv2d backprop to filter when not using cuDNN convolution
      • (since v2.7) tf.nn.dilation2d gradient
      • (since v2.7) tf.nn.max_pool_with_argmax gradient
      • (since v2.7) tf.raw_ops.DebugNumericSummary and tf.raw_ops.DebugNumericSummaryV2
      • (since v2.7) tf.timestamp. Throws FailedPrecondition
      • (since v2.7) tf.Variable.scatter_add (and other scatter methods, both on ref and resource variables)
      • (since v2.7) The random-number-generating ops in the tf.random module when the global random seed has not yet been set (via tf.random.set_seed). Throws RuntimeError from Python or InvalidArgument from C++
  • TensorFlow-oneDNN no longer supports explicit use of oneDNN blocked tensor format, e.g., setting the environment variable TF_ENABLE_MKL_NATIVE_FORMAT will not have any effect.

  • TensorFlow has been validated on Windows Subsystem for Linux 2 (aka WSL 2) for both GPUs and CPUs.

  • Due to security issues (see section below), all boosted trees code has been deprecated. Users should switch to TensorFlow Decision Forests. TF's boosted trees code will be eliminated before the branch cut for TF 2.9 and will no longer be present since that release.

Security

  • Fixes a floating point division by 0 when executing convolution operators (CVE-2022-21725)
  • Fixes a heap OOB read in shape inference for ReverseSequence (CVE-2022-21728)
  • Fixes a heap OOB access in Dequantize (CVE-2022-21726)
  • Fixes an integer overflow in shape inference for Dequantize (CVE-2022-21727)
  • Fixes a heap OOB access in FractionalAvgPoolGrad (CVE-2022-21730)
  • Fixes an overflow and divide by zero in UnravelIndex (CVE-2022-21729)
  • Fixes a type confusion in shape inference for ConcatV2 (CVE-2022-21731)
  • Fixes an OOM in ThreadPoolHandle (CVE-2022-21732)
  • Fixes an OOM due to integer overflow in StringNGrams (CVE-2022-21733)
  • Fixes more issues caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes an integer overflows in most sparse component-wise ops (CVE-2022-23567)
  • Fixes an integer overflows in AddManySparseToTensorsMap (CVE-2022-23568)
  • Fixes a number of CHECK-failures in MapStage (CVE-2022-21734)
  • Fixes a division by zero in FractionalMaxPool (CVE-2022-21735)
  • Fixes a number of CHECK-fails when building invalid/overflowing tensor shapes (CVE-2022-23569)
  • Fixes an undefined behavior in SparseTensorSliceDataset (CVE-2022-21736)
  • Fixes an assertion failure based denial of service via faulty bin count operations (CVE-2022-21737)
  • Fixes a reference binding to null pointer in QuantizedMaxPool (CVE-2022-21739)
  • Fixes an integer overflow leading to crash in SparseCountSparseOutput (CVE-2022-21738)
  • Fixes a heap overflow in SparseCountSparseOutput (CVE-2022-21740)
  • Fixes an FPE in BiasAndClamp in TFLite (CVE-2022-23557)
  • Fixes an FPE in depthwise convolutions in TFLite (CVE-2022-21741)
  • Fixes an integer overflow in TFLite array creation (CVE-2022-23558)
  • Fixes an integer overflow in TFLite (CVE-2022-23559)
  • Fixes a dangerous OOB write in TFLite (CVE-2022-23561)
  • Fixes a vulnerability leading to read and write outside of bounds in TFLite (CVE-2022-23560)
  • Fixes a set of vulnerabilities caused by using insecure temporary files (CVE-2022-23563)
  • Fixes an integer overflow in Range resulting in undefined behavior and OOM (CVE-2022-23562)
  • Fixes a vulnerability where missing validation causes tf.sparse.split to crash when axis is a tuple (CVE-2021-41206)
  • Fixes a CHECK-fail when decoding resource handles from proto (CVE-2022-23564)
  • Fixes a CHECK-fail with repeated AttrDef (CVE-2022-23565)
  • Fixes a heap OOB write in Grappler (CVE-2022-23566)
  • Fixes a CHECK-fail when decoding invalid tensors from proto (CVE-2022-23571)
  • Fixes a null-dereference when specializing tensor type (CVE-2022-23570)
  • Fixes a crash when type cannot be specialized (CVE-2022-23572)
  • Fixes a heap OOB read/write in SpecializeType (CVE-2022-23574)
  • Fixes an unitialized variable access in AssignOp (CVE-2022-23573)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateTensorSize (CVE-2022-23575)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateOutputSize (CVE-2022-23576)
  • Fixes a null dereference in GetInitOp (CVE-2022-23577)
  • Fixes a memory leak when a graph node is invalid (CVE-2022-23578)
  • Fixes an abort caused by allocating a vector that is too large (CVE-2022-23580)
  • Fixes multiple CHECK-failures during Grappler's IsSimplifiableReshape (CVE-2022-23581)
  • Fixes multiple CHECK-failures during Grappler's SafeToRemoveIdentity (CVE-2022-23579)
  • Fixes multiple CHECK-failures in TensorByteSize (CVE-2022-23582)
  • Fixes multiple CHECK-failures in binary ops due to type confusion (CVE-2022-23583)
  • Fixes a use after free in DecodePng kernel (CVE-2022-23584)
  • Fixes a memory leak in decoding PNG images (CVE-2022-23585)
  • Fixes multiple CHECK-fails in function.cc (CVE-2022-23586)
  • Fixes multiple CHECK-fails due to attempting to build a reference tensor (CVE-2022-23588)
  • Fixes an integer overflow in Grappler cost estimation of crop and resize operation (CVE-2022-23587)
  • Fixes a null pointer dereference in Grappler's IsConstant (CVE-2022-23589)
  • Fixes a CHECK failure in constant folding (CVE-2021-41197)
  • Fixes a stack overflow due to self-recursive function in GraphDef (CVE-2022-23591)
  • Fixes a heap OOB access in RunForwardTypeInference (CVE-2022-23592)
  • Fixes a crash due to erroneous StatusOr (CVE-2022-23590)
  • Fixes multiple crashes and heap OOB accesses in TFG dialect (MLIR) (CVE-2022-23594)
  • Fixes a segfault in simplifyBroadcast (MLIR) (CVE-2022-23593)
  • Fixes a null pointer dereference in BuildXlaCompilationCache (XLA) (CVE-2022-23595)
  • Updates icu to 69.1 to handle CVE-2020-10531

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

8bitmp3, Adam Lanicek, ag.ramesh, alesapin, Andrew Goodbody, annasuheyla, Ariel Elkin, Arnab Dutta, Ben Barsdell, bhack, cfRod, Chengji Yao, Christopher Bate, dan, Dan F-M, David Korczynski, DEKHTIARJonathan, dengzhiyuan, Deven Desai, Duncan Riach, Eli Osherovich, Ewout Ter Hoeven, ez2take, Faijul Amin, fo40225, Frederic Bastien, gadagashwini, Gauri1 Deshpande, Georgiy Manuilov, Guilherme De Lázari, Guozhong Zhuang, H1Gdev, homuler, Hongxu Jia, Jacky_Yin, jayfurmanek, jgehw, Jhalak Patel, Jinzhe Zeng, Johan Gunnarsson, Jonathan Dekhtiar, Kaixi Hou, Kanvi Khanna, Kevin Cheng, Koan-Sin Tan, Kruglov-Dmitry, Kun Lu, Lemo, Lequn Chen, long.chen, Louis Sugy, Mahmoud Abuzaina, Mao, Marius Brehler, Mark Harfouche, Martin Patz, Maxiwell S. Garcia, Meenakshi Venkataraman, Michael Melesse, Mrinal Tyagi, Måns Nilsson, Nathan John Sircombe, Nathan Luehr, Nilesh Agarwalla, Oktay Ozturk, Patrice Vignola, Pawel-Polyai, Rama Ketineni, Ramesh Sampath, Reza Rahimi, Rob Suderman, Robert Kalmar, Rohit Santhanam, Sachin Muradi, Saduf2019, Samuel Marks, Shi,Guangyong, Sidong-Wei, Srinivasan Narayanamoorthy, Srishti Srivastava, Steven I Reeves, stevenireeves, Supernovae, Tamas Bela Feher, Tao Xu, Thibaut Goetghebuer-Planchon, Thomas Schmeyer, tilakrayal, Valery Mironov, Victor Guo, Vignesh Kothapalli, Vishnuvardhan Janapati, wamuir, Wang,Quintin, William Muir, William Raveane, Yash Goel, Yimei Sun, Yong Tang, Yuduo Wu

Release 2.7.1

This releases introduces several vulnerability fixes:

  • Fixes a floating point division by 0 when executing convolution operators (CVE-2022-21725)
  • Fixes a heap OOB read in shape inference for ReverseSequence (CVE-2022-21728)
  • Fixes a heap OOB access in Dequantize (CVE-2022-21726)
  • Fixes an integer overflow in shape inference for Dequantize (CVE-2022-21727)
  • Fixes a heap OOB access in FractionalAvgPoolGrad (CVE-2022-21730)
  • Fixes an overflow and divide by zero in UnravelIndex (CVE-2022-21729)
  • Fixes a type confusion in shape inference for ConcatV2 (CVE-2022-21731)
  • Fixes an OOM in ThreadPoolHandle (CVE-2022-21732)
  • Fixes an OOM due to integer overflow in StringNGrams (CVE-2022-21733)
  • Fixes more issues caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes an integer overflows in most sparse component-wise ops (CVE-2022-23567)
  • Fixes an integer overflows in AddManySparseToTensorsMap (CVE-2022-23568)
  • Fixes a number of CHECK-failures in MapStage (CVE-2022-21734)
  • Fixes a division by zero in FractionalMaxPool (CVE-2022-21735)
  • Fixes a number of CHECK-fails when building invalid/overflowing tensor shapes (CVE-2022-23569)
  • Fixes an undefined behavior in SparseTensorSliceDataset (CVE-2022-21736)
  • Fixes an assertion failure based denial of service via faulty bin count operations (CVE-2022-21737)
  • Fixes a reference binding to null pointer in QuantizedMaxPool (CVE-2022-21739)
  • Fixes an integer overflow leading to crash in SparseCountSparseOutput (CVE-2022-21738)
  • Fixes a heap overflow in SparseCountSparseOutput (CVE-2022-21740)
  • Fixes an FPE in BiasAndClamp in TFLite (CVE-2022-23557)
  • Fixes an FPE in depthwise convolutions in TFLite (CVE-2022-21741)
  • Fixes an integer overflow in TFLite array creation (CVE-2022-23558)
  • Fixes an integer overflow in TFLite (CVE-2022-23559)
  • Fixes a dangerous OOB write in TFLite (CVE-2022-23561)
  • Fixes a vulnerability leading to read and write outside of bounds in TFLite (CVE-2022-23560)
  • Fixes a set of vulnerabilities caused by using insecure temporary files (CVE-2022-23563)
  • Fixes an integer overflow in Range resulting in undefined behavior and OOM (CVE-2022-23562)
  • Fixes a vulnerability where missing validation causes tf.sparse.split to crash when axis is a tuple (CVE-2021-41206)
  • Fixes a CHECK-fail when decoding resource handles from proto (CVE-2022-23564)
  • Fixes a CHECK-fail with repeated AttrDef (CVE-2022-23565)
  • Fixes a heap OOB write in Grappler (CVE-2022-23566)
  • Fixes a CHECK-fail when decoding invalid tensors from proto (CVE-2022-23571)
  • Fixes a null-dereference when specializing tensor type (CVE-2022-23570)
  • Fixes a crash when type cannot be specialized (CVE-2022-23572)
  • Fixes a heap OOB read/write in SpecializeType (CVE-2022-23574)
  • Fixes an uninitialized variable access in AssignOp (CVE-2022-23573)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateTensorSize (CVE-2022-23575)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateOutputSize (CVE-2022-23576)
  • Fixes a null dereference in GetInitOp (CVE-2022-23577)
  • Fixes a memory leak when a graph node is invalid (CVE-2022-23578)
  • Fixes an abort caused by allocating a vector that is too large (CVE-2022-23580)
  • Fixes multiple CHECK-failures during Grappler's IsSimplifiableReshape (CVE-2022-23581)
  • Fixes multiple CHECK-failures during Grappler's SafeToRemoveIdentity (CVE-2022-23579)
  • Fixes multiple CHECK-failures in TensorByteSize (CVE-2022-23582)
  • Fixes multiple CHECK-failures in binary ops due to type confusion (CVE-2022-23583)
  • Fixes a use after free in DecodePng kernel (CVE-2022-23584)
  • Fixes a memory leak in decoding PNG images (CVE-2022-23585)
  • Fixes multiple CHECK-fails in function.cc (CVE-2022-23586)
  • Fixes multiple CHECK-fails due to attempting to build a reference tensor (CVE-2022-23588)
  • Fixes an integer overflow in Grappler cost estimation of crop and resize operation (CVE-2022-23587)
  • Fixes a null pointer dereference in Grappler's IsConstant (CVE-2022-23589)
  • Fixes a CHECK failure in constant folding (CVE-2021-41197)
  • Fixes a stack overflow due to self-recursive function in GraphDef (CVE-2022-23591)
  • Fixes a crash due to erroneous StatusOr (CVE-2022-23590)
  • Fixes multiple crashes and heap OOB accesses in TFG dialect (MLIR) (CVE-2022-23594)
  • Fixes a null pointer dereference in BuildXlaCompilationCache (XLA) (CVE-2022-23595)
  • Updates icu to 69.1 to handle CVE-2020-10531

Release 2.6.3

This releases introduces several vulnerability fixes:

  • Fixes a floating point division by 0 when executing convolution operators (CVE-2022-21725)
  • Fixes a heap OOB read in shape inference for ReverseSequence (CVE-2022-21728)
  • Fixes a heap OOB access in Dequantize (CVE-2022-21726)
  • Fixes an integer overflow in shape inference for Dequantize (CVE-2022-21727)
  • Fixes a heap OOB access in FractionalAvgPoolGrad (CVE-2022-21730)
  • Fixes an overflow and divide by zero in UnravelIndex (CVE-2022-21729)
  • Fixes a type confusion in shape inference for ConcatV2 (CVE-2022-21731)
  • Fixes an OOM in ThreadPoolHandle (CVE-2022-21732)
  • Fixes an OOM due to integer overflow in StringNGrams (CVE-2022-21733)
  • Fixes more issues caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes an integer overflows in most sparse component-wise ops (CVE-2022-23567)
  • Fixes an integer overflows in AddManySparseToTensorsMap (CVE-2022-23568)
  • Fixes a number of CHECK-failures in MapStage (CVE-2022-21734)
  • Fixes a division by zero in FractionalMaxPool (CVE-2022-21735)
  • Fixes a number of CHECK-fails when building invalid/overflowing tensor shapes (CVE-2022-23569)
  • Fixes an undefined behavior in SparseTensorSliceDataset (CVE-2022-21736)
  • Fixes an assertion failure based denial of service via faulty bin count operations (CVE-2022-21737)
  • Fixes a reference binding to null pointer in QuantizedMaxPool (CVE-2022-21739)
  • Fixes an integer overflow leading to crash in SparseCountSparseOutput (CVE-2022-21738)
  • Fixes a heap overflow in SparseCountSparseOutput (CVE-2022-21740)
  • Fixes an FPE in BiasAndClamp in TFLite (CVE-2022-23557)
  • Fixes an FPE in depthwise convolutions in TFLite (CVE-2022-21741)
  • Fixes an integer overflow in TFLite array creation (CVE-2022-23558)
  • Fixes an integer overflow in TFLite (CVE-2022-23559)
  • Fixes a dangerous OOB write in TFLite (CVE-2022-23561)
  • Fixes a vulnerability leading to read and write outside of bounds in TFLite (CVE-2022-23560)
  • Fixes a set of vulnerabilities caused by using insecure temporary files (CVE-2022-23563)
  • Fixes an integer overflow in Range resulting in undefined behavior and OOM (CVE-2022-23562)
  • Fixes a vulnerability where missing validation causes tf.sparse.split to crash when axis is a tuple (CVE-2021-41206)
  • Fixes a CHECK-fail when decoding resource handles from proto (CVE-2022-23564)
  • Fixes a CHECK-fail with repeated AttrDef (CVE-2022-23565)
  • Fixes a heap OOB write in Grappler (CVE-2022-23566)
  • Fixes a CHECK-fail when decoding invalid tensors from proto (CVE-2022-23571)
  • Fixes a null-dereference when specializing tensor type (CVE-2022-23570)
  • Fixes a crash when type cannot be specialized (CVE-2022-23572)
  • Fixes a heap OOB read/write in SpecializeType (CVE-2022-23574)
  • Fixes an unitialized variable access in AssignOp (CVE-2022-23573)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateTensorSize (CVE-2022-23575)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateOutputSize (CVE-2022-23576)
  • Fixes a null dereference in GetInitOp (CVE-2022-23577)
  • Fixes a memory leak when a graph node is invalid (CVE-2022-23578)
  • Fixes an abort caused by allocating a vector that is too large (CVE-2022-23580)
  • Fixes multiple CHECK-failures during Grappler's IsSimplifiableReshape (CVE-2022-23581)
  • Fixes multiple CHECK-failures during Grappler's SafeToRemoveIdentity (CVE-2022-23579)
  • Fixes multiple CHECK-failures in TensorByteSize (CVE-2022-23582)
  • Fixes multiple CHECK-failures in binary ops due to type confusion (CVE-2022-23583)
  • Fixes a use after free in DecodePng kernel (CVE-2022-23584)
  • Fixes a memory leak in decoding PNG images (CVE-2022-23585)
  • Fixes multiple CHECK-fails in function.cc (CVE-2022-23586)
  • Fixes multiple CHECK-fails due to attempting to build a reference tensor (CVE-2022-23588)
  • Fixes an integer overflow in Grappler cost estimation of crop and resize operation (CVE-2022-23587)
  • Fixes a null pointer dereference in Grappler's IsConstant (CVE-2022-23589)
  • Fixes a CHECK failure in constant folding (CVE-2021-41197)
  • Fixes a stack overflow due to self-recursive function in GraphDef (CVE-2022-23591)
  • Fixes a null pointer dereference in BuildXlaCompilationCache (XLA) (CVE-2022-23595)
  • Updates icu to 69.1 to handle CVE-2020-10531

Release 2.5.3

This releases introduces several vulnerability fixes:

  • Fixes a floating point division by 0 when executing convolution operators (CVE-2022-21725)
  • Fixes a heap OOB read in shape inference for ReverseSequence (CVE-2022-21728)
  • Fixes a heap OOB access in Dequantize (CVE-2022-21726)
  • Fixes an integer overflow in shape inference for Dequantize (CVE-2022-21727)
  • Fixes a heap OOB access in FractionalAvgPoolGrad (CVE-2022-21730)
  • Fixes an overflow and divide by zero in UnravelIndex (CVE-2022-21729)
  • Fixes a type confusion in shape inference for ConcatV2 (CVE-2022-21731)
  • Fixes an OOM in ThreadPoolHandle (CVE-2022-21732)
  • Fixes an OOM due to integer overflow in StringNGrams (CVE-2022-21733)
  • Fixes more issues caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes an integer overflows in most sparse component-wise ops (CVE-2022-23567)
  • Fixes an integer overflows in AddManySparseToTensorsMap (CVE-2022-23568)
  • Fixes a number of CHECK-failures in MapStage (CVE-2022-21734)
  • Fixes a division by zero in FractionalMaxPool (CVE-2022-21735)
  • Fixes a number of CHECK-fails when building invalid/overflowing tensor shapes (CVE-2022-23569)
  • Fixes an undefined behavior in SparseTensorSliceDataset (CVE-2022-21736)
  • Fixes an assertion failure based denial of service via faulty bin count operations (CVE-2022-21737)
  • Fixes a reference binding to null pointer in QuantizedMaxPool (CVE-2022-21739)
  • Fixes an integer overflow leading to crash in SparseCountSparseOutput (CVE-2022-21738)
  • Fixes a heap overflow in SparseCountSparseOutput (CVE-2022-21740)
  • Fixes an FPE in BiasAndClamp in TFLite (CVE-2022-23557)
  • Fixes an FPE in depthwise convolutions in TFLite (CVE-2022-21741)
  • Fixes an integer overflow in TFLite array creation (CVE-2022-23558)
  • Fixes an integer overflow in TFLite (CVE-2022-23559)
  • Fixes a dangerous OOB write in TFLite (CVE-2022-23561)
  • Fixes a vulnerability leading to read and write outside of bounds in TFLite (CVE-2022-23560)
  • Fixes a set of vulnerabilities caused by using insecure temporary files (CVE-2022-23563)
  • Fixes an integer overflow in Range resulting in undefined behavior and OOM (CVE-2022-23562)
  • Fixes a vulnerability where missing validation causes tf.sparse.split to crash when axis is a tuple (CVE-2021-41206)
  • Fixes a CHECK-fail when decoding resource handles from proto (CVE-2022-23564)
  • Fixes a CHECK-fail with repeated AttrDef (CVE-2022-23565)
  • Fixes a heap OOB write in Grappler (CVE-2022-23566)
  • Fixes a CHECK-fail when decoding invalid tensors from proto (CVE-2022-23571)
  • Fixes an unitialized variable access in AssignOp (CVE-2022-23573)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateTensorSize (CVE-2022-23575)
  • Fixes an integer overflow in OpLevelCostEstimator::CalculateOutputSize (CVE-2022-23576)
  • Fixes a null dereference in GetInitOp (CVE-2022-23577)
  • Fixes a memory leak when a graph node is invalid (CVE-2022-23578)
  • Fixes an abort caused by allocating a vector that is too large (CVE-2022-23580)
  • Fixes multiple CHECK-failures during Grappler's IsSimplifiableReshape (CVE-2022-23581)
  • Fixes multiple CHECK-failures during Grappler's SafeToRemoveIdentity (CVE-2022-23579)
  • Fixes multiple CHECK-failures in TensorByteSize (CVE-2022-23582)
  • Fixes multiple CHECK-failures in binary ops due to type confusion (CVE-2022-23583)
  • Fixes a use after free in DecodePng kernel (CVE-2022-23584)
  • Fixes a memory leak in decoding PNG images (CVE-2022-23585)
  • Fixes multiple CHECK-fails in function.cc (CVE-2022-23586)
  • Fixes multiple CHECK-fails due to attempting to build a reference tensor (CVE-2022-23588)
  • Fixes an integer overflow in Grappler cost estimation of crop and resize operation (CVE-2022-23587)
  • Fixes a null pointer dereference in Grappler's IsConstant (CVE-2022-23589)
  • Fixes a CHECK failure in constant folding (CVE-2021-41197)
  • Fixes a stack overflow due to self-recursive function in GraphDef (CVE-2022-23591)
  • Updates icu to 69.1 to handle CVE-2020-10531

Release 2.7.0

Breaking Changes

  • tf.keras:

    • The methods Model.fit(), Model.predict(), and Model.evaluate() will no longer uprank input data of shape (batch_size,) to become (batch_size, 1). This enables Model subclasses to process scalar data in their train_step()/test_step()/predict_step() methods.
      Note that this change may break certain subclassed models. You can revert back to the previous behavior by adding upranking yourself in the train_step()/test_step()/predict_step() methods, e.g. if x.shape.rank == 1: x = tf.expand_dims(x, axis=-1). Functional models as well as Sequential models built with an explicit input shape are not affected.
    • The methods Model.to_yaml() and keras.models.model_from_yaml have been replaced to raise a RuntimeError as they can be abused to cause arbitrary code execution. It is recommended to use JSON serialization instead of YAML, or, a better alternative, serialize to H5.
    • LinearModel and WideDeepModel are moved to the tf.compat.v1.keras.models. namespace (tf.compat.v1.keras.models.LinearModel and tf.compat.v1.keras.models.WideDeepModel), and their experimental endpoints (tf.keras.experimental.models.LinearModel and tf.keras.experimental.models.WideDeepModel) are being deprecated.
    • RNG behavior change for all tf.keras.initializers classes. For any class constructed with a fixed seed, it will no longer generate same value when invoked multiple times. Instead, it will return different value, but a deterministic sequence. This change will make the initialize behavior align between v1 and v2.
  • tf.lite:

    • Rename fields SignatureDef table in schema to maximize the parity with TF SavedModel's Signature concept.
    • Deprecate Makefile builds. Makefile users need to migrate their builds to CMake or Bazel. Please refer to the Build TensorFlow Lite with CMake and Build TensorFlow Lite for ARM boards for the migration.
    • Deprecate tflite::OpResolver::GetDelegates. The list returned by TfLite's BuiltinOpResolver::GetDelegates is now always empty. Instead, recommend using new method tflite::OpResolver::GetDelegateCreators in order to achieve lazy initialization on TfLite delegate instances.
  • TF Core:

    • tf.Graph.get_name_scope() now always returns a string, as documented. Previously, when called within name_scope("") or name_scope(None) contexts, it returned None; now it returns the empty string.
    • tensorflow/core/ir/ contains a new MLIR-based Graph dialect that is isomorphic to GraphDef and will be used to replace GraphDef-based (e.g., Grappler) optimizations.
    • Deprecated and removed attrs() function in shape inference. All attributes should be queried by name now (rather than range returned) to enable changing the underlying storage there.
    • The following Python symbols were accidentally added in earlier versions of TensorFlow and now are removed. Each symbol has a replacement that should be used instead, but note the replacement's argument names are different.
      • tf.quantize_and_dequantize_v4 (accidentally introduced in TensorFlow 2.4): Use tf.quantization.quantize_and_dequantize_v2 instead.
      • tf.batch_mat_mul_v3 (accidentally introduced in TensorFlow 2.6): Use tf.linalg.matmul instead.
      • tf.sparse_segment_sum_grad (accidentally introduced in TensorFlow 2.6): Use tf.raw_ops.SparseSegmentSumGrad instead. Directly calling this op is typically not necessary, as it is automatically used when computing the gradient of tf.sparse.segment_sum.
    • Renaming of tensorflow::int64 to int_64_t in numerous places (the former is an alias for the latter) which could result in needing to regenerate selective op registration headers else execution would fail with unregistered kernels error.
  • Modular File System Migration:

    • Support for S3 and HDFS file systems has been migrated to a modular file systems based approach and is now available in https://github.com/tensorflow/io. The tensorflow-io python package should be installed for S3 and HDFS support with tensorflow.

Major Features and Improvements

  • Improvements to the TensorFlow debugging experience:

    • Previously, TensorFlow error stack traces involved many internal frames, which could be challenging to read through, while not being actionable for end users. As of TF 2.7, TensorFlow filters internal frames in most errors that it raises, to keep stack traces short, readable, and focused on what's actionable for end users (their own code).

    This behavior can be disabled by calling tf.debugging.disable_traceback_filtering(), and can be re-enabled via tf.debugging.enable_traceback_filtering(). If you are debugging a TensorFlow-internal issue (e.g. to prepare a TensorFlow PR), make sure to disable traceback filtering. You can check whether this feature is currently enabled by calling tf.debugging.is_traceback_filtering_enabled().

    Note that this feature is only available with Python 3.7 or higher.

    • Improve the informativeness of error messages raised by Keras Layer.__call__(), by adding the full list of argument values passed to the layer in every exception.
  • Introduce the tf.compat.v1.keras.utils.track_tf1_style_variables decorator, which enables using large classes of tf1-style variable_scope, get_variable, and compat.v1.layer-based components from within TF2 models running with TF2 behavior enabled.

  • tf.data:

    • tf.data service now supports auto-sharding. Users specify the sharding policy with tf.data.experimental.service.ShardingPolicy enum. It can be one of OFF (equivalent to today's "parallel_epochs" mode), DYNAMIC (equivalent to today's "distributed_epoch" mode), or one of the static sharding policies: FILE, DATA, FILE_OR_DATA, or HINT (corresponding to values of tf.data.experimental.AutoShardPolicy).

      Static sharding (auto-sharding) requires the number of tf.data service workers be fixed. Users need to specify the worker addresses in tensorflow.data.experimental.DispatcherConfig.

    • tf.data.experimental.service.register_dataset now accepts optional compression argument.

  • Keras:

    • tf.keras.layers.Conv now includes a public convolution_op method. This method can be used to simplify the implementation of Conv subclasses. There are two primary ways to use this new method. The first is to use the method directly in your own call method: python class StandardizedConv2D(tf.keras.layers.Conv2D): def call(self, inputs): mean, var = tf.nn.moments(self.kernel, axes=[0, 1, 2], keepdims=True) return self.convolution_op(inputs, (self.kernel - mean) / tf.sqrt(var + 1e-10)) Alternatively, you can override convolution_op: python class StandardizedConv2D(tf.keras.Layer): def convolution_op(self, inputs, kernel): mean, var = tf.nn.moments(kernel, axes=[0, 1, 2], keepdims=True) # Author code uses std + 1e-5 return super().convolution_op(inputs, (kernel - mean) / tf.sqrt(var + 1e-10))
    • Added merge_state() method to tf.keras.metrics.Metric for use in distributed computations.
    • Added sparse and ragged options to tf.keras.layers.TextVectorization to allow for SparseTensor and RaggedTensor outputs from the layer.
  • distribute.experimental.rpc package:

    • distribute.experimental.rpc package introduces APIs to create a GRPC based server to register tf.function methods and a GRPC client to invoke remote registered methods. RPC APIs are intended for multi-client setups i.e. server and clients are started in separate binaries independently.

    • Example usage to create server: ```python server = tf.distribute.experimental.rpc.Server.create("grpc", "127.0.0.1:1234") @tf.function(input_signature=[ tf.TensorSpec([], tf.int32), tf.TensorSpec([], dtypes.int32) ]) def _remote_multiply(a, b): return tf.math.multiply(a, b)

      server.register("multiply", _remote_multiply) ```

    • Example usage to create client: python client = tf.distribute.experimental.rpc.Client.create("grpc", address) a = tf.constant(2, dtype=tf.int32) b = tf.constant(3, dtype=tf.int32) result = client.multiply(a, b)

  • tf.lite:

    • Add experimental API experimental_from_jax to support conversion from Jax models to TensorFlow Lite.
    • Support uint32 data type for cast op.
    • Support int8 data type for cast op.
    • Add experimental quantization debugger tf.lite.QuantizationDebugger
    • Add lite.experimental.authoring.compatible API
      • A Python decorator to provide a way to check TFLite compatibility issue of tf.function. This returns a callable object which validates TFLite compatibility. If an incompatible operation is encountered during execution, an exception will be raised with information about the incompatible ops.
    • Add lite.experimental.Analyzer API
      • An experimental tool to analyze TFLite flatbuffer models. This API can be used to investigate TFLite model structure and check compatibility with GPU delegate.
  • Extension Types

    • Add experimental API to define new Python classes that can be handled by TensorFlow APIs. To create an extension type, simply define a Python class with tf.experimental.ExtensionType as its base, and use type annotations to specify the type for each field. E.g.: python class MaskedTensor(tf.experimental.ExtensionType): values: tf.Tensor mask: tf.Tensor The tf.ExtensionType base class works similarly to typing.NamedTuple and @dataclasses.dataclass from the standard Python library.
    • Extension types are supported by Keras, tf.data, TF-hub, SavedModel, tf.function, control flow ops, py_function, and distribution strategy.
    • Add "dispatch decorators" that can be used to override the default behavior of TensorFlow ops (such as tf.add or tf.concat) when they are applied to ExtensionType values.
    • The BatchableExtensionType API can be used to define extension types that support APIs that make use of batching, such as tf.data.Dataset and tf.map_fn.
    • For more information, see the Extension types guide.

Bug Fixes and Other Changes

  • TF Core:
    • Random number generation (RNG) system
      • Add argument alg to tf.random.stateless_* functions to explicitly select the RNG algorithm.
      • Add tf.nn.experimental.stateless_dropout, a stateless version of tf.nn.dropout.
      • tf.random.Generator now can be created inside the scope of tf.distribute.experimental.ParameterServerStrategy and tf.distribute.experimental.CentralStorageStrategy.
    • Add an experimental session config tf.experimental.disable_functional_ops_lowering which disables functional control flow op lowering optimization. This is useful when executing within a portable runtime where control flow op kernels may not be loaded due to selective registration.
    • Add a new experimental argument experimental_is_anonymous to tf.lookup.StaticHashTable.__init__ to create the table in anonymous mode. In this mode, the table resource can only be accessed via resource handles (not resource names) and will be deleted automatically when all resource handles pointing to it are gone.
  • tf.data:
    • Introduce the tf.data.experimental.at API which provides random access for input pipelines that consist of transformations that support random access. The initial set of transformations that support random access includes: tf.data.Dataset.from_tensor_slices,tf.data.Dataset.shuffle, tf.data.Dataset.batch, tf.data.Dataset.shard, tf.data.Dataset.map, and tf.data.Dataset.range.
    • Promote tf.data.Options.experimental_deterministic API to tf.data.Options.deterministic and deprecate the experimental endpoint.
    • Move autotuning options fromtf.data.Options.experimental_optimization.autotune* to a newly created tf.data.Options.autotune.* and remove support for tf.data.Options.experimental_optimization.autotune_buffers.
    • Add support for user-defined names of tf.data core Python API, which can be used to disambiguate tf.data events in TF Profiler Trace Viewer.
    • Promote tf.data.experimental.sample_from_datasets API to tf.data.Dataset.sample_from_datasets and deprecate the experimental endpoint.
    • Added TF_GPU_ALLOCATOR=cuda_malloc_async that use cudaMallocAsync from CUDA 11.2. This could become the default in the future.
  • TF SavedModel:
    • Custom gradients are now saved by default. See tf.saved_model.SaveOptions to disable this.
    • The saved_model_cli's --input_examples inputs are now restricted to python literals to avoid code injection.
  • XLA:
    • Add a new API that allows custom call functions to signal errors. The old API will be deprecated in a future release. See https://www.tensorflow.org/xla/custom_call for details.
    • XLA:GPU reductions are deterministic by default (reductions within jit_compile=True are now deterministic).
    • XLA:GPU works with Horovod (OSS contribution by Trent Lo from NVidia)
    • XLA:CPU and XLA:GPU can compile tf.unique and tf.where when shapes are provably correct at compile time.
  • tf.saved_model.save:
    • When saving a model, not specifying a namespace whitelist for custom ops with a namespace will now default to allowing rather than rejecting them all.
  • Deterministic Op Functionality (enabled by setting the environment variable TF_DETERMINISTIC_OPS to "true" or "1"):
    • Add determinsitic GPU implementations of:
      • tf.math.segment_sum
      • tf.math.segment_prod
      • tf.math.segment_mean
      • tf.math.unsorted_segment_sum
      • tf.math.unsorted_segment_prod
      • tf.math.unsorted_segment_sqrt
      • tf.math.unsorted_segment_mean
      • tf.gather backprop
      • tf.convert_to_tensor when fed with (sparse) tf.IndexedSlices
      • tf.nn.sparse_softmax_crossentropy_with_logits
      • tf.nn.ctc_loss (resolved, possibly in prior release, and confirmed with tests)
      • stateful ops used in tf.data.Dataset
    • Run the following ops on CPU (with significant performance penalty):
      • tf.scatter_nd and other related scatter functions, such as tf.tensor_scatter_nd_update
    • Add determinism-unimplemented exception-throwing to the following ops. When op-determinism is expected (i.e. when the environment variable TF_DETERMINISTIC_OPS is set to "true" or "1"), an attempt to use the specified paths through the following ops on a GPU will cause tf.errors.UnimplementedError (with an understandable message), unless otherwise specified, to be thrown.
      • tf.compat.v1.nn.fused_batch_norm backprop to offset when is_training=False
      • tf.image.adjust_contrast forward
      • tf.nn.depthwise_conv2d backprop to filter when not using cuDNN convolution
      • tf.image.resize with method=ResizeMethod.NEAREST backprop
      • tf.math.bincount - TODO: confirm exception added
      • tf.raw_ops.DebugNumericSummary and tf.raw_ops.DebugNumericSummaryV2
      • tf.Variable.scatter_add (and other scatter methods, both on ref and resource variables)
      • tf.linalg.svd
      • tf.nn.dilation2d gradient
      • tf.nn.max_pool_with_argmax gradient
      • tf.timestamp. Throws FailedPrecondition
      • The random-number-generating ops in the tf.random module when the global random seed has not yet been set (via tf.random.set_seed). Throws RuntimeError from Python or InvalidArgument from C++
      • tf.compat.v1.get_seed if the global random seed has not yet been set (via tf.random.set_seed). Throws RuntimeError from Python or InvalidArgument from C++

Security

  • Fixes a code injection issue in saved_model_cli (CVE-2021-41228)
  • Fixes a vulnerability due to use of uninitialized value in Tensorflow (CVE-2021-41225)
  • Fixes a heap OOB in FusedBatchNorm kernels (CVE-2021-41223)
  • Fixes an arbitrary memory read in ImmutableConst (CVE-2021-41227)
  • Fixes a heap OOB in SparseBinCount (CVE-2021-41226)
  • Fixes a heap OOB in SparseFillEmptyRows (CVE-2021-41224)
  • Fixes a segfault due to negative splits in SplitV (CVE-2021-41222)
  • Fixes segfaults and vulnerabilities caused by accesses to invalid memory during shape inference in Cudnn* ops (CVE-2021-41221)
  • Fixes a null pointer exception when Exit node is not preceded by Enter op (CVE-2021-41217)
  • Fixes an integer division by 0 in tf.raw_ops.AllToAll (CVE-2021-41218)
  • Fixes a use after free and a memory leak in CollectiveReduceV2 (CVE-2021-41220)
  • Fixes an undefined behavior via nullptr reference binding in sparse matrix multiplication (CVE-2021-41219)
  • Fixes a heap buffer overflow in Transpose (CVE-2021-41216)
  • Prevents deadlocks arising from mutually recursive tf.function objects (CVE-2021-41213)
  • Fixes a null pointer exception in DeserializeSparse (CVE-2021-41215)
  • Fixes an undefined behavior arising from reference binding to nullptr in tf.ragged.cross (CVE-2021-41214)
  • Fixes a heap OOB read in tf.ragged.cross (CVE-2021-41212)
  • Fixes a heap OOB in shape inference for QuantizeV2 (CVE-2021-41211)
  • Fixes a heap OOB read in all tf.raw_ops.QuantizeAndDequantizeV* ops (CVE-2021-41205)
  • Fixes an FPE in ParallelConcat (CVE-2021-41207)
  • Fixes FPE issues in convolutions with zero size filters (CVE-2021-41209)
  • Fixes a heap OOB read in tf.raw_ops.SparseCountSparseOutput (CVE-2021-41210)
  • Fixes vulnerabilities caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes vulnerabilities caused by incomplete validation of shapes in multiple TF ops (CVE-2021-41206)
  • Fixes a segfault produced while copying constant resource tensor (CVE-2021-41204)
  • Fixes a vulnerability caused by unitialized access in EinsumHelper::ParseEquation (CVE-2021-41201)
  • Fixes several vulnerabilities and segfaults caused by missing validation during checkpoint loading (CVE-2021-41203)
  • Fixes an overflow producing a crash in tf.range (CVE-2021-41202)
  • Fixes an overflow producing a crash in tf.image.resize when size is large (CVE-2021-41199)
  • Fixes an overflow producing a crash in tf.tile when tiling tensor is large (CVE-2021-41198)
  • Fixes a vulnerability produced due to incomplete validation in tf.summary.create_file_writer (CVE-2021-41200)
  • Fixes multiple crashes due to overflow and CHECK-fail in ops with large tensor shapes (CVE-2021-41197)
  • Fixes a crash in max_pool3d when size argument is 0 or negative (CVE-2021-41196)
  • Fixes a crash in tf.math.segment_* operations (CVE-2021-41195)
  • Updates curl to 7.78.0 to handle CVE-2021-22922, CVE-2021-22923, CVE-2021-22924, CVE-2021-22925, and CVE-2021-22926.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

8bitmp3, Abhilash Majumder, abhilash1910, AdeshChoudhar, Adrian Garcia Badaracco, Adrian Ratiu, ag.ramesh, Aleksandr Nikolaev, Alexander Bosch, Alexander Grund, Annie Tallund, Anush Elangovan, Artem Sokolovskii, azazhu, Balint Cristian, Bas Aarts, Ben Barsdell, bhack, cfRod, Cheney-Wang, Cheng Ren, Christopher Bate, collin, Danila Bespalov, David Datascientist, Deven Desai, Duncan Riach, Ehsan Kia, Ellie, Fan Du, fo40225, Frederic Bastien, fsx950223, Gauri1 Deshpande, geetachavan1, Guillaume Klein, guozhong.zhuang, helen, Håkon Sandsmark, japm48, jgehw, Jinzhe Zeng, Jonathan Dekhtiar, Kai Zhu, Kaixi Hou, Kanvi Khanna, Koan-Sin Tan, Koki Ibukuro, Kulin Seth, KumaTea, Kun-Lu, Lemo, lipracer, liuyuanqiang, Mahmoud Abuzaina, Marius Brehler, Maxiwell S. Garcia, mdfaijul, metarutaiga, Michal Szutenberg, nammbash, Neil Girdhar, Nishidha Panpaliya, Nyadla-Sys, Patrice Vignola, Peter Kasting, Philipp Hack, PINTO0309, Prateek Gupta, puneeshkhanna, Rahul Butani, Rajeshwar Reddy T, Reza Rahimi, RinozaJiffry, rmothukuru, Rohit Santhanam, Saduf2019, Samuel Marks, sclarkson, Sergii Khomenko, Sheng, Yang, Sidong-Wei, slowy07, Srinivasan Narayanamoorthy, Srishti Srivastava, stanley, Stella Alice Schlotter, Steven I Reeves, stevenireeves, svobora, Takayoshi Koizumi, Tamas Bela Feher, Thibaut Goetghebuer-Planchon, Trent Lo, Twice, Varghese, Jojimon, Vishnuvardhan Janapati, Wang Yanzhang, Wang,Quintin, William Muir, William Raveane, Yasir Modak, Yasuhiro Matsumoto, Yi Li, Yong Tang, zhaozheng09, Zhoulong Jiang, zzpmiracle

Release 2.6.2

Fixes an issue where keras, tensorflow_estimator and tensorboard were missing proper upper bounds and resulted in broken installs after TF 2.7 release

Release 2.6.1

This release introduces several vulnerability fixes:

  • Fixes a code injection issue in saved_model_cli (CVE-2021-41228)
  • Fixes a vulnerability due to use of uninitialized value in Tensorflow (CVE-2021-41225)
  • Fixes a heap OOB in FusedBatchNorm kernels (CVE-2021-41223)
  • Fixes an arbitrary memory read in ImmutableConst (CVE-2021-41227)
  • Fixes a heap OOB in SparseBinCount (CVE-2021-41226)
  • Fixes a heap OOB in SparseFillEmptyRows (CVE-2021-41224)
  • Fixes a segfault due to negative splits in SplitV (CVE-2021-41222)
  • Fixes segfaults and vulnerabilities caused by accesses to invalid memory during shape inference in Cudnn* ops (CVE-2021-41221)
  • Fixes a null pointer exception when Exit node is not preceded by Enter op (CVE-2021-41217)
  • Fixes an integer division by 0 in tf.raw_ops.AllToAll (CVE-2021-41218)
  • Fixes a use after free and a memory leak in CollectiveReduceV2 (CVE-2021-41220)
  • Fixes an undefined behavior via nullptr reference binding in sparse matrix multiplication (CVE-2021-41219)
  • Fixes a heap buffer overflow in Transpose (CVE-2021-41216)
  • Prevents deadlocks arising from mutually recursive tf.function objects (CVE-2021-41213)
  • Fixes a null pointer exception in DeserializeSparse (CVE-2021-41215)
  • Fixes an undefined behavior arising from reference binding to nullptr in tf.ragged.cross (CVE-2021-41214)
  • Fixes a heap OOB read in tf.ragged.cross (CVE-2021-41212)
  • Fixes a heap OOB in shape inference for QuantizeV2 (CVE-2021-41211)
  • Fixes a heap OOB read in all tf.raw_ops.QuantizeAndDequantizeV* ops (CVE-2021-41205)
  • Fixes an FPE in ParallelConcat (CVE-2021-41207)
  • Fixes FPE issues in convolutions with zero size filters (CVE-2021-41209)
  • Fixes a heap OOB read in tf.raw_ops.SparseCountSparseOutput (CVE-2021-41210)
  • Fixes vulnerabilities caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes vulnerabilities caused by incomplete validation of shapes in multiple TF ops (CVE-2021-41206)
  • Fixes a segfault produced while copying constant resource tensor (CVE-2021-41204)
  • Fixes a vulnerability caused by unitialized access in EinsumHelper::ParseEquation (CVE-2021-41201)
  • Fixes several vulnerabilities and segfaults caused by missing validation during checkpoint loading (CVE-2021-41203)
  • Fixes an overflow producing a crash in tf.range (CVE-2021-41202)
  • Fixes an overflow producing a crash in tf.image.resize when size is large (CVE-2021-41199)
  • Fixes an overflow producing a crash in tf.tile when tiling tensor is large (CVE-2021-41198)
  • Fixes a vulnerability produced due to incomplete validation in tf.summary.create_file_writer (CVE-2021-41200)
  • Fixes multiple crashes due to overflow and CHECK-fail in ops with large tensor shapes (CVE-2021-41197)
  • Fixes a crash in max_pool3d when size argument is 0 or negative (CVE-2021-41196)
  • Fixes a crash in tf.math.segment_* operations (CVE-2021-41195)
  • Updates curl to 7.78.0 to handle CVE-2021-22922, CVE-2021-22923, CVE-2021-22924, CVE-2021-22925, and CVE-2021-22926.

Release 2.6.0

Breaking Changes

  • tf.train.experimental.enable_mixed_precision_graph_rewrite is removed, as the API only works in graph mode and is not customizable. The function is still accessible under tf.compat.v1.mixed_precision.enable_mixed_precision_graph_rewrite, but it is recommended to use the Keras mixed precision API instead.

  • tf.lite:

    • Remove experimental.nn.dynamic_rnn, experimental.nn.TfLiteRNNCell and experimental.nn.TfLiteLSTMCell since they're no longer supported. It's recommended to just use keras lstm instead.
  • tf.keras:

    • Keras been split into a separate PIP package (keras), and its code has been moved to the GitHub repositorykeras-team/keras. The API endpoints for tf.keras stay unchanged, but are now backed by the keras PIP package. The existing code in tensorflow/python/keras is a staled copy and will be removed in future release (2.7). Please remove any imports to tensorflow.python.keras and replace them with public tf.keras API instead.
    • The methods Model.to_yaml() and keras.models.model_from_yaml have been replaced to raise a RuntimeError as they can be abused to cause arbitrary code execution. It is recommended to use JSON serialization instead of YAML, or, a better alternative, serialize to H5.

Known Caveats

  • TF Core:
    • A longstanding bug in tf.while_loop, which caused it to execute sequentially, even when parallel_iterations>1, has now been fixed. However, the increased parallelism may result in increased memory use. Users who experience unwanted regressions should reset their while_loop's parallel_iterations value to 1, which is consistent with prior behavior.

Major Features and Improvements

  • tf.keras:

    • Keras has been split into a separate PIP package (keras), and its code has been moved to the GitHub repository keras-team/keras. The API endpoints for tf.keras stay unchanged, but are now backed by the keras PIP package. All Keras-related PRs and issues should now be directed to the GitHub repository. keras-team/keras.
    • tf.keras.utils.experimental.DatasetCreator now takes an optional tf.distribute.InputOptions for specific options when used with distribution.
    • tf.keras.experimental.SidecarEvaluator is now available for a program intended to be run on an evaluator task, which is commonly used to supplement a training cluster running with tf.distribute.experimental.ParameterServerStrategy (see `https://www.tensorflow.org/tutorials/distribute/parameter_server_training). It can also be used with single-worker training or other strategies. See docstring for more info.
    • Preprocessing layers moved from experimental to core.
      • Import paths moved from tf.keras.layers.preprocessing.experimental to tf.keras.layers.
    • Updates to Preprocessing layers API for consistency and clarity:
      • StringLookup and IntegerLookup default for mask_token changed to None. This matches the default masking behavior of Hashing and Embedding layers. To keep existing behavior, pass mask_token="" during layer creation.
      • Renamed "binary" output mode to "multi_hot" for CategoryEncoding, StringLookup, IntegerLookup, and TextVectorization. Multi-hot encoding will no longer automatically uprank rank 1 inputs, so these layers can now multi-hot encode unbatched multi-dimensional samples.
      • Added a new output mode "one_hot" for CategoryEncoding, StringLookup, IntegerLookup, which will encode each element in an input batch individually, and automatically append a new output dimension if necessary. Use this mode on rank 1 inputs for the old "binary" behavior of one-hot encoding a batch of scalars.
      • Normalization will no longer automatically uprank rank 1 inputs, allowing normalization of unbatched multi-dimensional samples.
  • tf.lite:

    • The recommended Android NDK version for building TensorFlow Lite has been changed from r18b to r19c.
    • Supports int64 for mul.
    • Supports native variable builtin ops - ReadVariable, AssignVariable.
    • Converter:
      • Experimental support for variables in TFLite. To enable through conversion, users need to set experimental_enable_resource_variables on tf.lite.TFLiteConverter to True. Note: mutable variables is only available using from_saved_model in this release, support for other methods is coming soon.
      • Old Converter (TOCO) is getting removed from next release. It's been deprecated for few releases already.
  • tf.saved_model:

    • SavedModels can now save custom gradients. Use the option tf.saved_model.SaveOption(experimental_custom_gradients=True) to enable this feature. The documentation in Advanced autodiff has been updated.
    • Object metadata has now been deprecated and no longer saved to the SavedModel.
  • TF Core:

    • Added tf.config.experimental.reset_memory_stats to reset the tracked peak memory returned by tf.config.experimental.get_memory_info.
  • tf.data:

    • Added target_workers param to data_service_ops.from_dataset_id and data_service_ops.distribute. Users can specify "AUTO", "ANY", or "LOCAL" (case insensitive). If "AUTO", tf.data service runtime decides which workers to read from. If "ANY", TF workers read from any tf.data service workers. If "LOCAL", TF workers will only read from local in-processs tf.data service workers. "AUTO" works well for most cases, while users can specify other targets. For example, "LOCAL" would help avoid RPCs and data copy if every TF worker colocates with a tf.data service worker. Currently, "AUTO" reads from any tf.data service workers to preserve existing behavior. The default value is "AUTO".

Bug Fixes and Other Changes

  • TF Core:
    • Added tf.lookup.experimental.MutableHashTable, which provides a generic mutable hash table implementation.
      • Compared to tf.lookup.experimental.DenseHashTable this offers lower overall memory usage, and a cleaner API. It does not require specifying a delete_key and empty_key that cannot be inserted into the table.
    • Added support for specifying number of subdivisions in all reduce host collective. This parallelizes work on CPU and speeds up the collective performance. Default behavior is unchanged.
    • Add an option perturb_singular to tf.linalg.tridiagonal_solve that allows solving linear systems with a numerically singular tridiagonal matrix, e.g. for use in inverse iteration.
    • Added tf.linalg.eigh_tridiagonal that computes the eigenvalues of a Hermitian tridiagonal matrix.
    • tf.constant now places its output on the current default device.
    • SavedModel
      • Added tf.saved_model.experimental.TrackableResource, which allows the creation of custom wrapper objects for resource tensors.
      • Added a SavedModel load option to allow restoring partial checkpoints into the SavedModel. See tf.saved_model.LoadOptions for details.
    • Added a new op SparseSegmentSumGrad to match the other sparse segment gradient ops and avoid an extra gather operation that was in the previous gradient implementation.
    • Added a new session config setting internal_fragmentation_fraction, which controls when the BFC Allocator needs to split an oversized chunk to satisfy an allocation request.
    • Added tf.get_current_name_scope() which returns the current full name scope string that will be prepended to op names.
  • tf.data:
    • Promoting tf.data.experimental.bucket_by_sequence_length API to tf.data.Dataset.bucket_by_sequence_length and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.get_single_element API to tf.data.Dataset.get_single_element and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.group_by_window API to tf.data.Dataset.group_by_window and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.RandomDataset API to tf.data.Dataset.random and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.scan API to tf.data.Dataset.scan and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.snapshot API to tf.data.Dataset.shapshot and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.take_while API to tf.data.Dataset.take_while and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.ThreadingOptions API to tf.data.ThreadingOptions and deprecating the experimental endpoint.
    • Promoting tf.data.experimental.unique API to tf.data.Dataset.unique and deprecating the experimental endpoint.
    • Added stop_on_empty_dataset parameter to sample_from_datasets and choose_from_datasets. Setting stop_on_empty_dataset=True will stop sampling if it encounters an empty dataset. This preserves the sampling ratio throughout training. The prior behavior was to continue sampling, skipping over exhausted datasets, until all datasets are exhausted. By default, the original behavior (stop_on_empty_dataset=False) is preserved.
    • Removed previously deprecated tf.data statistics related APIs:
      • tf.data.Options.experimental_stats
      • tf.data.experimental.StatsAggregator
      • tf.data.experimental.StatsOptions.*
      • tf.data.experimental.bytes_produced_stats
      • tf.data.experimental.latency_stats
    • Removed the following experimental tf.data optimization APIs:
      • tf.data.experimental.MapVectorizationOptions.*
      • tf.data.experimental.OptimizationOptions.filter_with_random_uniform_fusion
      • tf.data.experimental.OptimizationOptions.hoist_random_uniform
      • tf.data.experimental.OptimizationOptions.map_vectorization * tf.data.experimental.OptimizationOptions.reorder_data_discarding_ops
  • tf.keras:
    • Fix usage of __getitem__ slicing in Keras Functional APIs when the inputs are RaggedTensor objects.
    • Add keepdims argument to all GlobalPooling layers.
    • Add include_preprocessing argument to MobileNetV3 architectures to control the inclusion of Rescaling layer in the model.
    • Add optional argument (force) to make_(train|test|predict)_funtion methods to skip the cached function and generate a new one. This is useful to regenerate in a single call the compiled training function when any .trainable attribute of any model's layer has changed.
    • Models now have a save_spec property which contains the TensorSpec specs for calling the model. This spec is automatically saved when the model is called for the first time.
  • tf.linalg:
    • Add CompositeTensor as a base class to LinearOperator.
  • tf.lite:
    • Fix mean op reference quantization rounding issue.
    • Added framework_stable BUILD target, which links in only the non-experimental TF Lite APIs.
    • Remove deprecated Java Interpreter methods:
      • modifyGraphWithDelegate - Use Interpreter.Options.addDelegate
      • setNumThreads - Use Interpreter.Options.setNumThreads
    • Add Conv3DTranspose as a builtin op.
  • tf.summary:
    • Fix tf.summary.should_record_summaries() so it correctly reflects when summaries will be written, even when tf.summary.record_if() is not n effect, by returning True tensor if default writer is present.
  • Grappler:
    • Disable default Grappler optimization timeout to make the optimization pipeline deterministic. This may lead to increased model loading time, because time spent in graph optimizations is now unbounded (was 20 minutes).
  • Deterministic Op Functionality (enabled by setting TF_DETERMINISTIC_OPS to "true" or "1"):
    • Add a deterministic GPU implementation of tf.nn.softmax_cross_entropy_with_logits. See PR 49178.
    • Add a deterministic CPU implementation of tf.image.crop_and_resize. See PR 48905.
    • Add determinism-unimplemented exception-throwing to the following ops. When op-determinism is expected, an attempt to use the specified paths through the following ops on a GPU will cause tf.errors.UnimplementedError (with an understandable message) to be thrown.
      • tf.nn.sparse_softmax_cross_entropy_with_logits forwards and/or backwards. See PR 47925.
      • tf.image.crop_and_resize gradient w.r.t. either image or boxes. See PR 48905.
      • tf.sparse.sparse_dense_matmul forwards. See PR 50355.

Security

  • Fixes a heap out of bounds access in sparse reduction operations (CVE-2021-37635)
  • Fixes a floating point exception in SparseDenseCwiseDiv (CVE-2021-37636)
  • Fixes a null pointer dereference in CompressElement (CVE-2021-37637)
  • Fixes a null pointer dereference in RaggedTensorToTensor (CVE-2021-37638)
  • Fixes a null pointer dereference and a heap OOB read arising from operations restoring tensors (CVE-2021-37639)
  • Fixes an integer division by 0 in sparse reshaping (CVE-2021-37640)
  • Fixes a division by 0 in ResourceScatterDiv (CVE-2021-37642)
  • Fixes a heap OOB in RaggedGather (CVE-2021-37641)
  • Fixes a std::abort raised from TensorListReserve (CVE-2021-37644)
  • Fixes a null pointer dereference in MatrixDiagPartOp (CVE-2021-37643)
  • Fixes an integer overflow due to conversion to unsigned (CVE-2021-37645)
  • Fixes a bad allocation error in StringNGrams caused by integer conversion (CVE-2021-37646)
  • Fixes a null pointer dereference in SparseTensorSliceDataset (CVE-2021-37647)
  • Fixes an incorrect validation of SaveV2 inputs (CVE-2021-37648)
  • Fixes a null pointer dereference in UncompressElement (CVE-2021-37649)
  • Fixes a segfault and a heap buffer overflow in {Experimental,}DatasetToTFRecord (CVE-2021-37650)
  • Fixes a heap buffer overflow in FractionalAvgPoolGrad (CVE-2021-37651)
  • Fixes a use after free in boosted trees creation (CVE-2021-37652)
  • Fixes a division by 0 in ResourceGather (CVE-2021-37653)
  • Fixes a heap OOB and a CHECK fail in ResourceGather (CVE-2021-37654)
  • Fixes a heap OOB in ResourceScatterUpdate (CVE-2021-37655)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToSparse (CVE-2021-37656)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixDiagV* ops (CVE-2021-37657)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixSetDiagV* ops (CVE-2021-37658)
  • Fixes an undefined behavior arising from reference binding to nullptr and heap OOB in binary cwise ops (CVE-2021-37659)
  • Fixes a division by 0 in inplace operations (CVE-2021-37660)
  • Fixes a crash caused by integer conversion to unsigned (CVE-2021-37661)
  • Fixes an undefined behavior arising from reference binding to nullptr in boosted trees (CVE-2021-37662)
  • Fixes a heap OOB in boosted trees (CVE-2021-37664)
  • Fixes vulnerabilities arising from incomplete validation in QuantizeV2 (CVE-2021-37663)
  • Fixes vulnerabilities arising from incomplete validation in MKL requantization (CVE-2021-37665)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToVariant (CVE-2021-37666)
  • Fixes an undefined behavior arising from reference binding to nullptr in unicode encoding (CVE-2021-37667)
  • Fixes an FPE in tf.raw_ops.UnravelIndex (CVE-2021-37668)
  • Fixes a crash in NMS ops caused by integer conversion to unsigned (CVE-2021-37669)
  • Fixes a heap OOB in UpperBound and LowerBound (CVE-2021-37670)
  • Fixes an undefined behavior arising from reference binding to nullptr in map operations (CVE-2021-37671)
  • Fixes a heap OOB in SdcaOptimizerV2 (CVE-2021-37672)
  • Fixes a CHECK-fail in MapStage (CVE-2021-37673)
  • Fixes a vulnerability arising from incomplete validation in MaxPoolGrad (CVE-2021-37674)
  • Fixes an undefined behavior arising from reference binding to nullptr in shape inference (CVE-2021-37676)
  • Fixes a division by 0 in most convolution operators (CVE-2021-37675)
  • Fixes vulnerabilities arising from missing validation in shape inference for Dequantize (CVE-2021-37677)
  • Fixes an arbitrary code execution due to YAML deserialization (CVE-2021-37678)
  • Fixes a heap OOB in nested tf.map_fn with RaggedTensors (CVE-2021-37679)
  • Fixes a division by zero in TFLite (CVE-2021-37680)
  • Fixes an NPE in TFLite (CVE-2021-37681)
  • Fixes a vulnerability arising from use of unitialized value in TFLite (CVE-2021-37682)
  • Fixes an FPE in TFLite division operations (CVE-2021-37683)
  • Fixes an FPE in TFLite pooling operations (CVE-2021-37684)
  • Fixes an infinite loop in TFLite (CVE-2021-37686)
  • Fixes a heap OOB in TFLite (CVE-2021-37685)
  • Fixes a heap OOB in TFLite's Gather* implementations (CVE-2021-37687)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite (CVE-2021-37688)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite MLIR optimizations (CVE-2021-37689)
  • Fixes a FPE in LSH in TFLite (CVE-2021-37691)
  • Fixes a segfault on strings tensors with mismatched dimensions, arising in Go code (CVE-2021-37692)
  • Fixes a use after free and a potential segfault in shape inference functions (CVE-2021-37690)
  • Updates curl to 7.77.0 to handle CVE-2021-22876, CVE-2021-22897, CVE-2021-22898, and CVE-2021-22901.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Aadhitya A, Abhilash Mahendrakar, Abhishek Varma, Abin Shahab, Adam Hillier, Aditya Kane, AdityaKane2001, ag.ramesh, Amogh Joshi, Armen Poghosov, armkevincheng, Avrosh K, Ayan Moitra, azazhu, Banikumar Maiti, Bas Aarts, bhack, Bhanu Prakash Bandaru Venkata, Billy Cao, Bohumir Zamecnik, Bradley Reece, CyanXu, Daniel Situnayake, David Pal, Ddavis-2015, DEKHTIARJonathan, Deven Desai, Duncan Riach, Edward, Eli Osherovich, Eugene Kuznetsov, europeanplaice, evelynmitchell, Evgeniy Polyakov, Felix Vollmer, Florentin Hennecker, François Chollet, Frederic Bastien, Fredrik Knutsson, Gabriele Macchi, Gaurav Shukla, Gauri1 Deshpande, geetachavan1, Georgiy Manuilov, H, Hengwen Tong, Henri Woodcock, Hiran Sarkar, Ilya Arzhannikov, Janghoo Lee, jdematos, Jens Meder, Jerry Shih, jgehw, Jim Fisher, Jingbei Li, Jiri Podivin, Joachim Gehweiler, Johannes Lade, Jonas I. Liechti, Jonas Liechti, Jonas Ohlsson, Jonathan Dekhtiar, Julian Gross, Kaixi Hou, Kevin Cheng, Koan-Sin Tan, Kulin Seth, linzewen, Liubov Batanina, luisleee, Lukas Geiger, Mahmoud Abuzaina, mathgaming, Matt Conley, Max H. Gerlach, mdfaijul, Mh Kwon, Michael Martis, Michal Szutenberg, Måns Nilsson, nammbash, Neil Girdhar, Nicholas Vadivelu, Nick Kreeger, Nirjas Jakilim, okyanusoz, Patrice Vignola, Patrik Laurell, Pedro Marques, Philipp Hack, Phillip Cloud, Piergiacomo De Marchi, Prashant Kumar, puneeshkhanna, pvarouktsis, QQ喵, Rajeshwar Reddy T, Rama Ketineni, Reza Rahimi, Robert Kalmar, rsun, Ryan Kuester, Saduf2019, Sean Morgan, Sean Moriarity, Shaochen Shi, Sheng, Yang, Shu Wang, Shuai Zhang, Soojeong, Stanley-Nod, Steven I Reeves, stevenireeves, Suraj Sudhir, Sven Mayer, Tamas Bela Feher, tashuang.zk, tcervi, Teng Lu, Thales Elero Cervi, Thibaut Goetghebuer-Planchon, Thomas Walther, Till Brychcy, Trent Lo, Uday Bondhugula, vishakha.agrawal, Vishnuvardhan Janapati, wamuir, Wenwen Ouyang, wenwu, Williard Joshua Jose, xiaohong1031, Xiaoming (Jason) Cui, Xinan Jiang, Yasir Modak, Yi Li, Yong Tang, zilinzhu, 박상준, 이장

Release 2.5.2

This release introduces several vulnerability fixes:

  • Fixes a code injection issue in saved_model_cli (CVE-2021-41228)
  • Fixes a vulnerability due to use of uninitialized value in Tensorflow (CVE-2021-41225)
  • Fixes a heap OOB in FusedBatchNorm kernels (CVE-2021-41223)
  • Fixes an arbitrary memory read in ImmutableConst (CVE-2021-41227)
  • Fixes a heap OOB in SparseBinCount (CVE-2021-41226)
  • Fixes a heap OOB in SparseFillEmptyRows (CVE-2021-41224)
  • Fixes a segfault due to negative splits in SplitV (CVE-2021-41222)
  • Fixes segfaults and vulnerabilities caused by accesses to invalid memory during shape inference in Cudnn* ops (CVE-2021-41221)
  • Fixes a null pointer exception when Exit node is not preceded by Enter op (CVE-2021-41217)
  • Fixes an integer division by 0 in tf.raw_ops.AllToAll (CVE-2021-41218)
  • Fixes an undefined behavior via nullptr reference binding in sparse matrix multiplication (CVE-2021-41219)
  • Fixes a heap buffer overflow in Transpose (CVE-2021-41216)
  • Prevents deadlocks arising from mutually recursive tf.function objects (CVE-2021-41213)
  • Fixes a null pointer exception in DeserializeSparse (CVE-2021-41215)
  • Fixes an undefined behavior arising from reference binding to nullptr in tf.ragged.cross (CVE-2021-41214)
  • Fixes a heap OOB read in tf.ragged.cross (CVE-2021-41212)
  • Fixes a heap OOB read in all tf.raw_ops.QuantizeAndDequantizeV* ops (CVE-2021-41205)
  • Fixes an FPE in ParallelConcat (CVE-2021-41207)
  • Fixes FPE issues in convolutions with zero size filters (CVE-2021-41209)
  • Fixes a heap OOB read in tf.raw_ops.SparseCountSparseOutput (CVE-2021-41210)
  • Fixes vulnerabilities caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes vulnerabilities caused by incomplete validation of shapes in multiple TF ops (CVE-2021-41206)
  • Fixes a segfault produced while copying constant resource tensor (CVE-2021-41204)
  • Fixes a vulnerability caused by unitialized access in EinsumHelper::ParseEquation (CVE-2021-41201)
  • Fixes several vulnerabilities and segfaults caused by missing validation during checkpoint loading (CVE-2021-41203)
  • Fixes an overflow producing a crash in tf.range (CVE-2021-41202)
  • Fixes an overflow producing a crash in tf.image.resize when size is large (CVE-2021-41199)
  • Fixes an overflow producing a crash in tf.tile when tiling tensor is large (CVE-2021-41198)
  • Fixes a vulnerability produced due to incomplete validation in tf.summary.create_file_writer (CVE-2021-41200)
  • Fixes multiple crashes due to overflow and CHECK-fail in ops with large tensor shapes (CVE-2021-41197)
  • Fixes a crash in max_pool3d when size argument is 0 or negative (CVE-2021-41196)
  • Fixes a crash in tf.math.segment_* operations (CVE-2021-41195)
  • Updates curl to 7.78.0 to handle CVE-2021-22922, CVE-2021-22923, CVE-2021-22924, CVE-2021-22925, and CVE-2021-22926.

Release 2.5.1

This release introduces several vulnerability fixes:

  • Fixes a heap out of bounds access in sparse reduction operations (CVE-2021-37635)
  • Fixes a floating point exception in SparseDenseCwiseDiv (CVE-2021-37636)
  • Fixes a null pointer dereference in CompressElement (CVE-2021-37637)
  • Fixes a null pointer dereference in RaggedTensorToTensor (CVE-2021-37638)
  • Fixes a null pointer dereference and a heap OOB read arising from operations restoring tensors (CVE-2021-37639)
  • Fixes an integer division by 0 in sparse reshaping (CVE-2021-37640)
  • Fixes a division by 0 in ResourceScatterDiv (CVE-2021-37642)
  • Fixes a heap OOB in RaggedGather (CVE-2021-37641)
  • Fixes a std::abort raised from TensorListReserve (CVE-2021-37644)
  • Fixes a null pointer dereference in MatrixDiagPartOp (CVE-2021-37643)
  • Fixes an integer overflow due to conversion to unsigned (CVE-2021-37645)
  • Fixes a bad allocation error in StringNGrams caused by integer conversion (CVE-2021-37646)
  • Fixes a null pointer dereference in SparseTensorSliceDataset (CVE-2021-37647)
  • Fixes an incorrect validation of SaveV2 inputs (CVE-2021-37648)
  • Fixes a null pointer dereference in UncompressElement (CVE-2021-37649)
  • Fixes a segfault and a heap buffer overflow in {Experimental,}DatasetToTFRecord (CVE-2021-37650)
  • Fixes a heap buffer overflow in FractionalAvgPoolGrad (CVE-2021-37651)
  • Fixes a use after free in boosted trees creation (CVE-2021-37652)
  • Fixes a division by 0 in ResourceGather (CVE-2021-37653)
  • Fixes a heap OOB and a CHECK fail in ResourceGather (CVE-2021-37654)
  • Fixes a heap OOB in ResourceScatterUpdate (CVE-2021-37655)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToSparse (CVE-2021-37656)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixDiagV* ops (CVE-2021-37657)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixSetDiagV* ops (CVE-2021-37658)
  • Fixes an undefined behavior arising from reference binding to nullptr and heap OOB in binary cwise ops (CVE-2021-37659)
  • Fixes a division by 0 in inplace operations (CVE-2021-37660)
  • Fixes a crash caused by integer conversion to unsigned (CVE-2021-37661)
  • Fixes an undefined behavior arising from reference binding to nullptr in boosted trees (CVE-2021-37662)
  • Fixes a heap OOB in boosted trees (CVE-2021-37664)
  • Fixes vulnerabilities arising from incomplete validation in QuantizeV2 (CVE-2021-37663)
  • Fixes vulnerabilities arising from incomplete validation in MKL requantization (CVE-2021-37665)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToVariant (CVE-2021-37666)
  • Fixes an undefined behavior arising from reference binding to nullptr in unicode encoding (CVE-2021-37667)
  • Fixes an FPE in tf.raw_ops.UnravelIndex (CVE-2021-37668)
  • Fixes a crash in NMS ops caused by integer conversion to unsigned (CVE-2021-37669)
  • Fixes a heap OOB in UpperBound and LowerBound (CVE-2021-37670)
  • Fixes an undefined behavior arising from reference binding to nullptr in map operations (CVE-2021-37671)
  • Fixes a heap OOB in SdcaOptimizerV2 (CVE-2021-37672)
  • Fixes a CHECK-fail in MapStage (CVE-2021-37673)
  • Fixes a vulnerability arising from incomplete validation in MaxPoolGrad (CVE-2021-37674)
  • Fixes an undefined behavior arising from reference binding to nullptr in shape inference (CVE-2021-37676)
  • Fixes a division by 0 in most convolution operators (CVE-2021-37675)
  • Fixes vulnerabilities arising from missing validation in shape inference for Dequantize (CVE-2021-37677)
  • Fixes an arbitrary code execution due to YAML deserialization (CVE-2021-37678)
  • Fixes a heap OOB in nested tf.map_fn with RaggedTensors (CVE-2021-37679)
  • Fixes a division by zero in TFLite (CVE-2021-37680)
  • Fixes an NPE in TFLite (CVE-2021-37681)
  • Fixes a vulnerability arising from use of unitialized value in TFLite (CVE-2021-37682)
  • Fixes an FPE in TFLite division operations (CVE-2021-37683)
  • Fixes an FPE in TFLite pooling operations (CVE-2021-37684)
  • Fixes an infinite loop in TFLite (CVE-2021-37686)
  • Fixes a heap OOB in TFLite (CVE-2021-37685)
  • Fixes a heap OOB in TFLite's Gather* implementations (CVE-2021-37687)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite (CVE-2021-37688)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite MLIR optimizations (CVE-2021-37689)
  • Fixes a FPE in LSH in TFLite (CVE-2021-37691)
  • Fixes a segfault on strings tensors with mismatched dimensions, arising in Go code (CVE-2021-37692)
  • Fixes a use after free and a potential segfault in shape inference functions (CVE-2021-37690)
  • Updates curl to 7.77.0 to handle CVE-2021-22876, CVE-2021-22897, CVE-2021-22898, and CVE-2021-22901.

Release 2.4.4

This release introduces several vulnerability fixes:

  • Fixes a code injection issue in saved_model_cli (CVE-2021-41228)
  • Fixes a vulnerability due to use of uninitialized value in Tensorflow (CVE-2021-41225)
  • Fixes a heap OOB in FusedBatchNorm kernels (CVE-2021-41223)
  • Fixes an arbitrary memory read in ImmutableConst (CVE-2021-41227)
  • Fixes a heap OOB in SparseBinCount (CVE-2021-41226)
  • Fixes a heap OOB in SparseFillEmptyRows (CVE-2021-41224)
  • Fixes a segfault due to negative splits in SplitV (CVE-2021-41222)
  • Fixes segfaults and vulnerabilities caused by accesses to invalid memory during shape inference in Cudnn* ops (CVE-2021-41221)
  • Fixes a null pointer exception when Exit node is not preceded by Enter op (CVE-2021-41217)
  • Fixes an integer division by 0 in tf.raw_ops.AllToAll (CVE-2021-41218)
  • Fixes an undefined behavior via nullptr reference binding in sparse matrix multiplication (CVE-2021-41219)
  • Fixes a heap buffer overflow in Transpose (CVE-2021-41216)
  • Prevents deadlocks arising from mutually recursive tf.function objects (CVE-2021-41213)
  • Fixes a null pointer exception in DeserializeSparse (CVE-2021-41215)
  • Fixes an undefined behavior arising from reference binding to nullptr in tf.ragged.cross (CVE-2021-41214)
  • Fixes a heap OOB read in tf.ragged.cross (CVE-2021-41212)
  • Fixes a heap OOB read in all tf.raw_ops.QuantizeAndDequantizeV* ops (CVE-2021-41205)
  • Fixes an FPE in ParallelConcat (CVE-2021-41207)
  • Fixes FPE issues in convolutions with zero size filters (CVE-2021-41209)
  • Fixes a heap OOB read in tf.raw_ops.SparseCountSparseOutput (CVE-2021-41210)
  • Fixes vulnerabilities caused by incomplete validation in boosted trees code (CVE-2021-41208)
  • Fixes vulnerabilities caused by incomplete validation of shapes in multiple TF ops (CVE-2021-41206)
  • Fixes a segfault produced while copying constant resource tensor (CVE-2021-41204)
  • Fixes a vulnerability caused by unitialized access in EinsumHelper::ParseEquation (CVE-2021-41201)
  • Fixes several vulnerabilities and segfaults caused by missing validation during checkpoint loading (CVE-2021-41203)
  • Fixes an overflow producing a crash in tf.range (CVE-2021-41202)
  • Fixes an overflow producing a crash in tf.image.resize when size is large (CVE-2021-41199)
  • Fixes an overflow producing a crash in tf.tile when tiling tensor is large (CVE-2021-41198)
  • Fixes a vulnerability produced due to incomplete validation in tf.summary.create_file_writer (CVE-2021-41200)
  • Fixes multiple crashes due to overflow and CHECK-fail in ops with large tensor shapes (CVE-2021-41197)
  • Fixes a crash in max_pool3d when size argument is 0 or negative (CVE-2021-41196)
  • Fixes a crash in tf.math.segment_* operations (CVE-2021-41195)
  • Updates curl to 7.78.0 to handle CVE-2021-22922, CVE-2021-22923, CVE-2021-22924, CVE-2021-22925, and CVE-2021-22926.

Release 2.4.3

This release introduces several vulnerability fixes:

  • Fixes a heap out of bounds access in sparse reduction operations (CVE-2021-37635)
  • Fixes a floating point exception in SparseDenseCwiseDiv (CVE-2021-37636)
  • Fixes a null pointer dereference in CompressElement (CVE-2021-37637)
  • Fixes a null pointer dereference in RaggedTensorToTensor (CVE-2021-37638)
  • Fixes a null pointer dereference and a heap OOB read arising from operations restoring tensors (CVE-2021-37639)
  • Fixes an integer division by 0 in sparse reshaping (CVE-2021-37640)
  • Fixes a division by 0 in ResourceScatterDiv (CVE-2021-37642)
  • Fixes a heap OOB in RaggedGather (CVE-2021-37641)
  • Fixes a std::abort raised from TensorListReserve (CVE-2021-37644)
  • Fixes a null pointer dereference in MatrixDiagPartOp (CVE-2021-37643)
  • Fixes an integer overflow due to conversion to unsigned (CVE-2021-37645)
  • Fixes a bad allocation error in StringNGrams caused by integer conversion (CVE-2021-37646)
  • Fixes a null pointer dereference in SparseTensorSliceDataset (CVE-2021-37647)
  • Fixes an incorrect validation of SaveV2 inputs (CVE-2021-37648)
  • Fixes a null pointer dereference in UncompressElement (CVE-2021-37649)
  • Fixes a segfault and a heap buffer overflow in {Experimental,}DatasetToTFRecord (CVE-2021-37650)
  • Fixes a heap buffer overflow in FractionalAvgPoolGrad (CVE-2021-37651)
  • Fixes a use after free in boosted trees creation (CVE-2021-37652)
  • Fixes a division by 0 in ResourceGather (CVE-2021-37653)
  • Fixes a heap OOB and a CHECK fail in ResourceGather (CVE-2021-37654)
  • Fixes a heap OOB in ResourceScatterUpdate (CVE-2021-37655)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToSparse (CVE-2021-37656)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixDiagV* ops (CVE-2021-37657)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixSetDiagV* ops (CVE-2021-37658)
  • Fixes an undefined behavior arising from reference binding to nullptr and heap OOB in binary cwise ops (CVE-2021-37659)
  • Fixes a division by 0 in inplace operations (CVE-2021-37660)
  • Fixes a crash caused by integer conversion to unsigned (CVE-2021-37661)
  • Fixes an undefined behavior arising from reference binding to nullptr in boosted trees (CVE-2021-37662)
  • Fixes a heap OOB in boosted trees (CVE-2021-37664)
  • Fixes vulnerabilities arising from incomplete validation in QuantizeV2 (CVE-2021-37663)
  • Fixes vulnerabilities arising from incomplete validation in MKL requantization (CVE-2021-37665)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToVariant (CVE-2021-37666)
  • Fixes an undefined behavior arising from reference binding to nullptr in unicode encoding (CVE-2021-37667)
  • Fixes an FPE in tf.raw_ops.UnravelIndex (CVE-2021-37668)
  • Fixes a crash in NMS ops caused by integer conversion to unsigned (CVE-2021-37669)
  • Fixes a heap OOB in UpperBound and LowerBound (CVE-2021-37670)
  • Fixes an undefined behavior arising from reference binding to nullptr in map operations (CVE-2021-37671)
  • Fixes a heap OOB in SdcaOptimizerV2 (CVE-2021-37672)
  • Fixes a CHECK-fail in MapStage (CVE-2021-37673)
  • Fixes a vulnerability arising from incomplete validation in MaxPoolGrad (CVE-2021-37674)
  • Fixes an undefined behavior arising from reference binding to nullptr in shape inference (CVE-2021-37676)
  • Fixes a division by 0 in most convolution operators (CVE-2021-37675)
  • Fixes vulnerabilities arising from missing validation in shape inference for Dequantize (CVE-2021-37677)
  • Fixes an arbitrary code execution due to YAML deserialization (CVE-2021-37678)
  • Fixes a heap OOB in nested tf.map_fn with RaggedTensors (CVE-2021-37679)
  • Fixes a division by zero in TFLite (CVE-2021-37680)
  • Fixes an NPE in TFLite (CVE-2021-37681)
  • Fixes a vulnerability arising from use of unitialized value in TFLite (CVE-2021-37682)
  • Fixes an FPE in TFLite division operations (CVE-2021-37683)
  • Fixes an FPE in TFLite pooling operations (CVE-2021-37684)
  • Fixes an infinite loop in TFLite (CVE-2021-37686)
  • Fixes a heap OOB in TFLite (CVE-2021-37685)
  • Fixes a heap OOB in TFLite's Gather* implementations (CVE-2021-37687)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite (CVE-2021-37688)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite MLIR optimizations (CVE-2021-37689)
  • Fixes a FPE in LSH in TFLite (CVE-2021-37691)
  • Fixes a segfault on strings tensors with mismatched dimensions, arising in Go code (CVE-2021-37692)
  • Fixes a use after free and a potential segfault in shape inference functions (CVE-2021-37690)
  • Updates curl to 7.77.0 to handle CVE-2021-22876, CVE-2021-22897, CVE-2021-22898, and CVE-2021-22901.

Release 2.3.4

This release introduces several vulnerability fixes:

  • Fixes a heap out of bounds access in sparse reduction operations (CVE-2021-37635)
  • Fixes a floating point exception in SparseDenseCwiseDiv (CVE-2021-37636)
  • Fixes a null pointer dereference in CompressElement (CVE-2021-37637)
  • Fixes a null pointer dereference in RaggedTensorToTensor (CVE-2021-37638)
  • Fixes a null pointer dereference and a heap OOB read arising from operations restoring tensors (CVE-2021-37639)
  • Fixes an integer division by 0 in sparse reshaping (CVE-2021-37640)
  • Fixes a division by 0 in ResourceScatterDiv (CVE-2021-37642)
  • Fixes a heap OOB in RaggedGather (CVE-2021-37641)
  • Fixes a std::abort raised from TensorListReserve (CVE-2021-37644)
  • Fixes a null pointer dereference in MatrixDiagPartOp (CVE-2021-37643)
  • Fixes an integer overflow due to conversion to unsigned (CVE-2021-37645)
  • Fixes a bad allocation error in StringNGrams caused by integer conversion (CVE-2021-37646)
  • Fixes a null pointer dereference in SparseTensorSliceDataset (CVE-2021-37647)
  • Fixes an incorrect validation of SaveV2 inputs (CVE-2021-37648)
  • Fixes a null pointer dereference in UncompressElement (CVE-2021-37649)
  • Fixes a segfault and a heap buffer overflow in {Experimental,}DatasetToTFRecord (CVE-2021-37650)
  • Fixes a heap buffer overflow in FractionalAvgPoolGrad (CVE-2021-37651)
  • Fixes a use after free in boosted trees creation (CVE-2021-37652)
  • Fixes a division by 0 in ResourceGather (CVE-2021-37653)
  • Fixes a heap OOB and a CHECK fail in ResourceGather (CVE-2021-37654)
  • Fixes a heap OOB in ResourceScatterUpdate (CVE-2021-37655)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToSparse (CVE-2021-37656)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixDiagV* ops (CVE-2021-37657)
  • Fixes an undefined behavior arising from reference binding to nullptr in MatrixSetDiagV* ops (CVE-2021-37658)
  • Fixes an undefined behavior arising from reference binding to nullptr and heap OOB in binary cwise ops (CVE-2021-37659)
  • Fixes a division by 0 in inplace operations (CVE-2021-37660)
  • Fixes a crash caused by integer conversion to unsigned (CVE-2021-37661)
  • Fixes an undefined behavior arising from reference binding to nullptr in boosted trees (CVE-2021-37662)
  • Fixes a heap OOB in boosted trees (CVE-2021-37664)
  • Fixes vulnerabilities arising from incomplete validation in QuantizeV2 (CVE-2021-37663)
  • Fixes vulnerabilities arising from incomplete validation in MKL requantization (CVE-2021-37665)
  • Fixes an undefined behavior arising from reference binding to nullptr in RaggedTensorToVariant (CVE-2021-37666)
  • Fixes an undefined behavior arising from reference binding to nullptr in unicode encoding (CVE-2021-37667)
  • Fixes an FPE in tf.raw_ops.UnravelIndex (CVE-2021-37668)
  • Fixes a crash in NMS ops caused by integer conversion to unsigned (CVE-2021-37669)
  • Fixes a heap OOB in UpperBound and LowerBound (CVE-2021-37670)
  • Fixes an undefined behavior arising from reference binding to nullptr in map operations (CVE-2021-37671)
  • Fixes a heap OOB in SdcaOptimizerV2 (CVE-2021-37672)
  • Fixes a CHECK-fail in MapStage (CVE-2021-37673)
  • Fixes a vulnerability arising from incomplete validation in MaxPoolGrad (CVE-2021-37674)
  • Fixes an undefined behavior arising from reference binding to nullptr in shape inference (CVE-2021-37676)
  • Fixes a division by 0 in most convolution operators (CVE-2021-37675)
  • Fixes vulnerabilities arising from missing validation in shape inference for Dequantize (CVE-2021-37677)
  • Fixes an arbitrary code execution due to YAML deserialization (CVE-2021-37678)
  • Fixes a heap OOB in nested tf.map_fn with RaggedTensors (CVE-2021-37679)
  • Fixes a division by zero in TFLite (CVE-2021-37680)
  • Fixes an NPE in TFLite (CVE-2021-37681)
  • Fixes a vulnerability arising from use of unitialized value in TFLite (CVE-2021-37682)
  • Fixes an FPE in TFLite division operations (CVE-2021-37683)
  • Fixes an FPE in TFLite pooling operations (CVE-2021-37684)
  • Fixes an infinite loop in TFLite (CVE-2021-37686)
  • Fixes a heap OOB in TFLite (CVE-2021-37685)
  • Fixes a heap OOB in TFLite's Gather* implementations (CVE-2021-37687)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite (CVE-2021-37688)
  • Fixes an undefined behavior arising from null pointer dereference in TFLite MLIR optimizations (CVE-2021-37689)
  • Fixes a FPE in LSH in TFLite (CVE-2021-37691)
  • Fixes a segfault on strings tensors with mismatched dimensions, arising in Go code (CVE-2021-37692)
  • Fixes a use after free and a potential segfault in shape inference functions (CVE-2021-37690)
  • Updates curl to 7.77.0 to handle CVE-2021-22876, CVE-2021-22897, CVE-2021-22898, and CVE-2021-22901.

Release 2.4.2

This release introduces several vulnerability fixes:

Release 2.3.3

This release introduces several vulnerability fixes:

Release 2.2.3

This release introduces several vulnerability fixes:

Release 2.1.4

This release introduces several vulnerability fixes:

Release 2.5.0

Major Features and Improvements

  • Support for Python3.9 has been added.
  • tf.data:
    • tf.data service now supports strict round-robin reads, which is useful for synchronous training workloads where example sizes vary. With strict round robin reads, users can guarantee that consumers get similar-sized examples in the same step.
    • tf.data service now supports optional compression. Previously data would always be compressed, but now you can disable compression by passing compression=None to tf.data.experimental.service.distribute(...).
    • tf.data.Dataset.batch() now supports num_parallel_calls and deterministic arguments. num_parallel_calls is used to indicate that multiple input batches should be computed in parallel. With num_parallel_calls set, deterministic is used to indicate that outputs can be obtained in the non-deterministic order.
    • Options returned by tf.data.Dataset.options() are no longer mutable.
    • tf.data input pipelines can now be executed in debug mode, which disables any asynchrony, parallelism, or non-determinism and forces Python execution (as opposed to trace-compiled graph execution) of user-defined functions passed into transformations such as map. The debug mode can be enabled through tf.data.experimental.enable_debug_mode().
  • tf.lite
    • Enabled the new MLIR-based quantization backend by default
      • The new backend is used for 8 bits full integer post-training quantization
      • The new backend removes the redundant rescales and fixes some bugs (shared weight/bias, extremely small scales, etc)
      • Set experimental_new_quantizer in tf.lite.TFLiteConverter to False to disable this change
  • tf.keras
    • tf.keras.metrics.AUC now support logit predictions.
    • Enabled a new supported input type in Model.fit, tf.keras.utils.experimental.DatasetCreator, which takes a callable, dataset_fn. DatasetCreator is intended to work across all tf.distribute strategies, and is the only input type supported for Parameter Server strategy.
  • tf.distribute
    • tf.distribute.experimental.ParameterServerStrategy now supports training with Keras Model.fit when used with DatasetCreator.
    • Creating tf.random.Generator under tf.distribute.Strategy scopes is now allowed (except for tf.distribute.experimental.CentralStorageStrategy and tf.distribute.experimental.ParameterServerStrategy). Different replicas will get different random-number streams.
  • TPU embedding support
    • Added profile_data_directory to EmbeddingConfigSpec in _tpu_estimator_embedding.py. This allows embedding lookup statistics gathered at runtime to be used in embedding layer partitioning decisions.
  • PluggableDevice
  • oneAPI Deep Neural Network Library (oneDNN) CPU performance optimizations from Intel-optimized TensorFlow are now available in the official x86-64 Linux and Windows builds.
    • They are off by default. Enable them by setting the environment variable TF_ENABLE_ONEDNN_OPTS=1.
    • We do not recommend using them in GPU systems, as they have not been sufficiently tested with GPUs yet.
  • TensorFlow pip packages are now built with CUDA11.2 and cuDNN 8.1.0

Breaking Changes

  • The TF_CPP_MIN_VLOG_LEVEL environment variable has been renamed to TF_CPP_MAX_VLOG_LEVEL which correctly describes its effect.

Bug Fixes and Other Changes

  • tf.keras:

    • Preprocessing layers API consistency changes:
      • StringLookup added output_mode, sparse, and pad_to_max_tokens arguments with same semantics as TextVectorization.
      • IntegerLookup added output_mode, sparse, and pad_to_max_tokens arguments with same semantics as TextVectorization. Renamed max_values, oov_value and mask_value to max_tokens, oov_token and mask_token to align with StringLookup and TextVectorization.
      • TextVectorization default for pad_to_max_tokens switched to False.
      • CategoryEncoding no longer supports adapt, IntegerLookup now supports equivalent functionality. max_tokens argument renamed to num_tokens.
      • Discretization added num_bins argument for learning bins boundaries through calling adapt on a dataset. Renamed bins argument to bin_boundaries for specifying bins without adapt.
    • Improvements to model saving/loading:
      • model.load_weights now accepts paths to saved models.
    • Keras inputs can now be created directly from arbitrary tf.TypeSpecs.
    • Two new learning rate schedules added: tf.keras.optimizers.schedules.CosineDecay andtf.keras.optimizers.schedules.CosineDecayRestarts.
  • tf.data:

    • Exposing tf.data.experimental.ExternalStatePolicy, which can be used to control how external state should be handled during dataset serialization or iterator checkpointing.
    • Changing tf.data.experimental.save to store the type specification of the dataset elements. This avoids the need for explicitly specifying the element_spec argument of tf.data.experimental.load when loading the previously saved dataset.
    • Add .element_spec property to tf.data.DatasetSpec to access the inner spec. This can be used to extract the structure of nested datasets.
    • Add tf.data.experimental.AutoShardingPolicy.HINT which can be used to provide hints to tf.distribute-based auto-sharding as to where in the input pipeline to insert sharding transformations.
    • Make tf.data.Options persistent across tf.function and GraphDef boundaries.
  • XLA compilation:

    • tf.function(experimental_compile=True) has become a stable API, renamed tf.function(jit_compile=True).
    • XLA can now compile MirroredStrategy: the step function passed tostrategy.run can now be annoted with jit_compile=True.
  • tf.distribute:

    • Rename experimental_prefetch_to_device in tf.distribute.InputOptions to experimental_fetch_to_device to better reflect the purpose.
  • tf.lite:

    • class tflite::Subgraph:
      • Removed the tensors() method and the non-const overload of the nodes_and_registration() method, both of which were previously documented as temporary and to be removed.
        • Uses of tensors() can be replaced by calling the existing methods tensors_size() and tensor(int).
        • Uses of the non-const overload of nodes_and_registration can be replaced by calling the existing methods nodes_size() and context(), and then calling the GetNodeAndRegistration method in the TfLiteContext returned by context().
    • NNAPI
      • Removed deprecated Interpreter::UseNNAPI(bool) C++ API.
        • Use NnApiDelegate() and related delegate configuration methods directly.
      • Replaced the model cache key for models computation algorithm with one guaranteed to be stable across runs.
    • 16 bits quantization
      • Added int16x8 support for ABS, REDUCE_MAX and REDUCE_MIN operators.
      • Additional tests and fixes for ADD and SUB operators.
    • Added support for saved model's session initializer through TFLiteConverter.from_saved_model.
    • Added DEPTH_TO_SPACE support in Post training quantization.
    • Added dynamic range quantization support for the BatchMatMul op.
      • Both symmetric and asymmetric quantized input tensor are supported.
    • Add RFFT2D as builtin op. (RFFT2D also supports RFFTD.) Currently only supports float32 input.
    • Add 5D support to SLICE op.
    • TFLite Supports SingatureDef:
      • TFLiteConverter exports models with SignatureDef
      • Interpreter supports getting a list of signatures and getting callable function for a given signaturedef.
    • Add int8 support for ReshapeV2.
    • Add experimental support for optimization with sparsity.
    • Add nominal support for unsigned 32-bit integer tensor types. Note that very few TFLite kernels support this type natively, so its use in mobile ML authoring is generally discouraged.
    • Add support for static hash tables through TFLiteConverter.from_saved_model.
    • The Python TF Lite Interpreter bindings now has an option experimental_preserve_all_tensors to aid in debugging conversion.
    • Quantized x86 execution defaults to Ruy GEMM library for platforms with AVX support.
    • Deprecate tf.compat.v1.lite.experimental.get_potentially_supported_ops. Use tf.lite.TFLiteConverter directly to check whether a model is convertible.
    • Add support to select one of three different built-in op resolvers
    • Enabled post training with calibrations for models that require user provided TensorFlow Lite custom op libraries via converter.target_spec._experimental_custom_op_registerers. used in Python Interpreter API.
  • TF Core:

    • Corrected higher-order gradients of control flow constructs (tf.cond, tf.while_loop, and compositions like tf.foldl) computed with tf.GradientTape inside a tf.function.
    • Changed the default step size in gradient_checker_v2.compute_gradients to be exactly representable as a binary floating point numbers. This avoids poluting gradient approximations needlessly, which is some cases leads to false negatives in op gradient tests.
    • Added tf.config.experimental.get_memory_info, returning a dict with the current and peak memory usage. Deprecated tf.config.experimental.get_memory_usage in favor of this new function.
    • Extended tf.config.experimental.enable_tensor_float_32_execution to control Tensor-Float-32 evaluation in RNNs.
    • Added a 'experimental_payloads' field to tf.errors.OpError and its subclasses to support more detailed error reporting. This is inspired from Abseil Status payloads: https://github.com/abseil/abseil-cpp/blob/master/absl/status/status.h
  • tf.summary:

    • New tf.summary.graph allows manual write of TensorFlow graph (tf.Graph or tf.compat.v1.GraphDef) as a summary. This is not a replacement for the trace-based API.
  • Set /d2ReducedOptimizeHugeFunctions by default for Windows builds. This provides a big compile-time speedup, and effectively raises the minimum supported MSVC version to 16.4 (current: 16.8).

  • TensorRT

    • Removed the deprecated session_config parameter for the TF1-TRT converter TrtGraphConverter. Previously, we issued a warning when the value of the parameter is not None.
    • The TF2-TRT converter TrtGraphConverterV2 takes an object of class TrtConversionParams as a parameter. Removed three deprecated fields from this class: rewriter_config_template, is_dynamic_op, and max_batch_size. Previously, we issued a warning when the value of rewriter_config_template is not None. We issued an error when the value of is_dynamic_op is not True. We didn't use the value for max_batch_size for building TensorRT engines. Add parameters use_dynamic_shape to enable dynamic shape support. The default is to disable dynamic shape support. Add dynamic_shape_profile_strategy for selecting a dynamic shape profile strategy. The default is profile strategy is Range.
    • Issue a warning when function get_tensorrt_rewriter_config is used.
  • TF XLA

    • Add new enum value MLIR_BRIDGE_ROLLOUT_SAFE_MODE_ENABLED to tf.config.experimental.mlir_bridge_rollout to enable a "safe" mode. This runs the MLIR bridge only when an analysis of the graph only when an analysis of the graph determines that it is safe to run.
    • Add new enum value MLIR_BRIDGE_ROLLOUT_SAFE_MODE_FALLBACK_ENABLED' totf.config.experimental.mlir_bridge_rollout` to enable a fallback for the MLIR bridge in a "safe" mode. This runs the MLIR bridge in a FallbackEnabled mode when an analysis of the graph determines that the graph does not have unsupported features.
  • Deterministic Op Functionality:

    • Add determinism-unimplemented exception-throwing to the segment-sum ops. When the environment variable TF_DETERMINISTIC_OPS is set to "true" or "1" (when op-determinism is expected), an attempt to run the following ops on a GPU will throw tf.errors.UnimplementedError (with an understandable message) when data is a floating-point type, including complex types (if supported): tf.math.segment_prod, tf.math.segment_sum, tf.math.unsorted_segment_mean, tf.math.unsorted_segment_sqrt_n, tf.math.unsorted_segment_prod, tf.math.unsorted_segment_sum, and therefore also tf.convert_to_tensor when value is of type tf.IndexedSlices (such as in the back prop though tf.gather into a dense embedding). See issue 39751 which this change addresses, but does not solve. This exception-throwing behavior can be disabled by setting the environment variable TF_DISABLE_SEGMENT_REDUCTION_OP_DETERMINISM_EXCEPTIONS to "true" or "1". For more information about these changes, see the description in pull request 47772.
    • In previous versions of TensorFlow, when a GPU was available, tf.sparse.sparse_dense_matmul introduced truly random noise in the forward path for data of type tf.float32 but not for data of type tf.float64 (for which there was no GPU implementation). In this current release, GPU support for other floating-point types (tf.float16, tf.float64, tf.complex64, and tf.complex128) has been added for this op. If you were relying on the determinism of the tf.float64 CPU implementation being automatically selected because of the absence of the tf.float64 GPU implementation, you with either need to force the op to run on the CPU or use a different data type.
  • Security

  • Other

    • Added show_debug_info to mlir.convert_graph_def and mlir.convert_function.
    • Added Arm Compute Library (ACL) support to --config=mkl_aarch64 build.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

8bitmp3, Aaron S. Mondal, Abhilash Mahendrakar, Abhinav Upadhyay, Abhishek Kulkarni, Abolfazl Shahbazi, Adam Hillier, Aditya Kane, Ag Ramesh, ahmedsabie, Albert Villanova Del Moral, Aleksey Vitebskiy, Alex Hoffman, Alexander Bayandin, Alfie Edwards, Aman Kishore, Amogh Joshi, andreABbauer, Andrew Goodbody, Andrzej Pomirski, Artemiy Ryabinkov, Ashish Jha, ather, Ayan Moitra, Bairen Yi, Bart Ribbers, Bas Aarts, Behzad Abghari, Ben Arnao, Ben Barsdell, Benjamin Klimczak, bhack, Brendan Collins, Can Wang, Cheng Ren, Chris Leary, Chris Olivier, Clemens Giuliani, Cloud Han, Corey Cole, Cui, Yifeng, Cuong V. Nguyen, Daniel Moore, Dawid Wojciechowski, Ddavis-2015, Dean Wyatte, Denisa Roberts, dependabot[bot], Dmitry Volodin, Dominic Jack, Duncan Riach, dushuai, Elena Zhelezina, Eli Osherovich, Erik Smistad, ewsn1593, Felix Fent, fo40225, François Chollet, Frederic Bastien, Freedom" Koan-Sin Tan, fsx950223, ganand1, gbaned, Georgiy Manuilov, gerbauz, Guillaume Klein, Guozhong Zhuang, Harry Slatyer, Harsh188, henri, Henri Woodcock, Hiran Sarkar, Hollow Man, Håkon Sandsmark, I Wayan Dharmana, icysapphire, Ikko Ashimine, Jab Hofmeier, Jack Hessel, Jacob Valdez, Jakub Jatczak, James Bernardi, Jared Smolens, Jason Zaman, jedlimlx, Jenny Plunkett, Jens Elofsson, Jerry Shih, jgehw, Jia Fu Low, Jim Fisher, jpodivin, Julien Stephan, Jungsub Lim, Junha Park, Junhyuk So, justkw, Kaixi Hou, kashyapraval, Kasra Bigdeli, Kazuaki Ishizaki, Keith Mok, Kevin Cheng, kopytjuk, Kristian Hartikainen, ksood12345, Kulin Seth, kushanam, latyas, Lequn Chen, Leslie-Fang, Long M. Lưu, Lukas Geiger, machineko, Mahmoud Abuzaina, Manish, Mao Yunfei, Maozhou, Ge, Marcin Juszkiewicz, Marcin Owsiany, Marconi Jiang, Marcos Pereira, Maria Romanenko Vexlard, Maria Vexlard, Marius Brehler, marload, Martin Kubovčík, Matej, Mateusz Holenko, Maxiwell S. Garcia, Mazhar, mazharul, mbhuiyan, mdfaijul, Michael Gielda, Michael Kuchnik, Michal Szutenberg, Mikhail Stepanov, Milan Straka, Mitchel Humpherys, Mohamed Moselhy, Mohamed Nour Abouelseoud, Måns Bermell, Måns Nilsson, Nathan Luehr, Nico Jahn, Niroop Ammbashankar, Oceania2018, Omri Steiner, Orivej Desh, Oskar Flordal, oujiafan, Patrik Laurell, Paul B. Isaac'S, Paul Klinger, Pawel Piskorski, Pedro Marques, Phat Tran, Piotr Zierhoffer, piyushdatta, Pnikam-Cad, Prashant Kumar, Prateek Gupta, PratsBhatt, Pravin Karandikar, qqq.jq, QQ喵, Quintin, Rama Ketineni, ravikyram, Rehan Guha, rhdong, rmothukuru, Roger Cheng, Rohit Santhanam, rposts, Rsanthanam-Amd, rsun, Rsun-Bdti, Ryan Kuester, ryanking13, Saduf2019, Sami Kama, Samuel Marks, Scott Tseng, Sean Moriarity, Sergey Popov, Sergii Khomenko, Sheng, Yang, shwetaoj, Sidong-Wei, Simon Maurer, Simrit Kaur, Srini511, Srinivasan Narayanamoorthy, Stephan, Stephen Matthews, Sungmann Cho, Sunoru, Suraj Sudhir, Suraj Upadhyay, Taebum Kim, Takayoshi Koizumi, Tamas Bela Feher, Teng Lu, Thibaut Goetghebuer-Planchon, Tomwildenhain-Microsoft, Tony, Traun Leyden, Trent Lo, TVLIgnacy, Tzu-Wei Sung, vaibhav, Vignesh Kothapalli, Vikram Dattu, viktprog, Vinayaka Bandishti, Vincent Abriou, Vishakha Agrawal, Vivek Panyam, Vladimir Silyaev, Võ Văn Nghĩa, wamuir, Wang, Yanzhang, wangsiyu, Waqar Hameed, wxinix, Xiao Yang, xiaohong1031, Xiaoming (Jason) Cui, Xinan Jiang, Yair Ehrenwald, Yajush Vyas, Yasir Modak, Yimei Sun, Yong Tang, Yosshi999, youshenmebutuo, yqtianust, Yuan Tang, yuanbopeng, Yuriy Chernyshov, Yuta Fukasawa, Zachary Deane-Mayer, Zeno Gantner, Zhoulong Jiang, zhuyie, zilinzhu, 彭震东

Release 2.4.1

  • This release removes the AVX2 requirement from TF 2.4.0.

Release 2.3.2

Bug Fixes and Other Changes

  • Fixes an access to unitialized memory in Eigen code (CVE-2020-26266)
  • Fixes a security vulnerability caused by lack of validation in tf.raw_ops.DataFormatVecPermute and tf.raw_ops.DataFormatDimMap (CVE-2020-26267)
  • Fixes a vulnerability caused by attempting to write to immutable memory region in tf.raw_ops.ImmutableConst (CVE-2020-26268
  • Fixes a CHECK-fail in LSTM with zero-length input (CVE-2020-26270)
  • Fixes a security vulnerability caused by accessing heap data outside of bounds when loading a specially crafted SavedModel (CVE-2020-26271)
  • Solves an OOM issue on TPUs when XLA contexts use fused average updates
  • Updates libjpeg-turbo to 2.0.5 to handle CVE-2020-13790.
  • Updates junit to 4.13.1 to handle CVE-2020-15250.
  • Updates PCRE to 8.44 to handle CVE-2019-20838 and CVE-2020-14155.
  • Updates sqlite3 to 3.44.0 to keep in sync with master branch.

Release 2.2.2

Bug Fixes and Other Changes

  • Fixes an access to unitialized memory in Eigen code (CVE-2020-26266)
  • Fixes a security vulnerability caused by lack of validation in tf.raw_ops.DataFormatVecPermute and tf.raw_ops.DataFormatDimMap (CVE-2020-26267)
  • Fixes a vulnerability caused by attempting to write to immutable memory region in tf.raw_ops.ImmutableConst (CVE-2020-26268
  • Fixes a CHECK-fail in LSTM with zero-length input (CVE-2020-26270)
  • Fixes a security vulnerability caused by accessing heap data outside of bounds when loading a specially crafted SavedModel (CVE-2020-26271)
  • Prevents memory leaks in loading SavedModels that import functions
  • Updates libjpeg-turbo to 2.0.5 to handle CVE-2020-13790.
  • Updates junit to 4.13.1 to handle CVE-2020-15250.
  • Updates PCRE to 8.44 to handle CVE-2019-20838 and CVE-2020-14155.
  • Updates sqlite3 to 3.44.0 to keep in sync with master branch.

Release 2.1.3

Bug Fixes and Other Changes

  • Fixes an access to unitialized memory in Eigen code (CVE-2020-26266)
  • Fixes a security vulnerability caused by lack of validation in tf.raw_ops.DataFormatVecPermute and tf.raw_ops.DataFormatDimMap (CVE-2020-26267)
  • Fixes a vulnerability caused by attempting to write to immutable memory region in tf.raw_ops.ImmutableConst (CVE-2020-26268
  • Fixes a CHECK-fail in LSTM with zero-length input (CVE-2020-26270)
  • Fixes a security vulnerability caused by accessing heap data outside of bounds when loading a specially crafted SavedModel (CVE-2020-26271)
  • Updates libjpeg-turbo to 2.0.5 to handle CVE-2020-13790.
  • Updates junit to 4.13.1 to handle CVE-2020-15250.
  • Updates PCRE to 8.44 to handle CVE-2019-20838 and CVE-2020-14155.
  • Updates sqlite3 to 3.44.0 to keep in sync with master branch.
  • Newer ROCm versions are supported on the 2.1 branch.

Release 2.0.4

Note that this is the last patch release for the TensorFlow 2.0.x series.

Bug Fixes and Other Changes

  • Fixes an access to unitialized memory in Eigen code (CVE-2020-26266)
  • Fixes a security vulnerability caused by lack of validation in tf.raw_ops.DataFormatVecPermute and tf.raw_ops.DataFormatDimMap (CVE-2020-26267)
  • Fixes a vulnerability caused by attempting to write to immutable memory region in tf.raw_ops.ImmutableConst (CVE-2020-26268
  • Fixes a CHECK-fail in LSTM with zero-length input (CVE-2020-26270)
  • Fixes a security vulnerability caused by accessing heap data outside of bounds when loading a specially crafted SavedModel (CVE-2020-26271)
  • Updates libjpeg-turbo to 2.0.5 to handle CVE-2020-13790.
  • Updates junit to 4.13.1 to handle CVE-2020-15250.
  • Updates PCRE to 8.44 to handle CVE-2019-20838 and CVE-2020-14155.
  • Updates sqlite3 to 3.44.0 to keep in sync with master branch.

Release 1.15.5

Note that this is the last patch release for the TensorFlow 1.x series.

Bug Fixes and Other Changes

  • Fixes an access to unitialized memory in Eigen code (CVE-2020-26266)
  • Fixes a security vulnerability caused by lack of validation in tf.raw_ops.DataFormatVecPermute and tf.raw_ops.DataFormatDimMap (CVE-2020-26267)
  • Fixes a vulnerability caused by attempting to write to immutable memory region in tf.raw_ops.ImmutableConst (CVE-2020-26268
  • Fixes a CHECK-fail in LSTM with zero-length input (CVE-2020-26270)
  • Fixes a security vulnerability caused by accessing heap data outside of bounds when loading a specially crafted SavedModel (CVE-2020-26271)
  • Updates libjpeg-turbo to 2.0.5 to handle CVE-2020-13790.
  • Updates junit to 4.13.1 to handle CVE-2020-15250.
  • Updates PCRE to 8.44 to handle CVE-2019-20838 and CVE-2020-14155.
  • Updates sqlite3 to 3.44.0 to keep in sync with master branch.

Release 2.4.0

## Major Features and Improvements

  • tf.distribute introduces experimental support for asynchronous training of models via the tf.distribute.experimental.ParameterServerStrategy API. Please see the tutorial to learn more.

  • MultiWorkerMirroredStrategy is now a stable API and is no longer considered experimental. Some of the major improvements involve handling peer failure and many bug fixes. Please check out the detailed tutorial on Multi-worker training with Keras.

  • Introduces experimental support for a new module named tf.experimental.numpy which is a NumPy-compatible API for writing TF programs. See the detailed guide to learn more. Additional details below.

  • Adds Support for TensorFloat-32 on Ampere based GPUs. TensorFloat-32, or TF32 for short, is a math mode for NVIDIA Ampere based GPUs and is enabled by default.

  • A major refactoring of the internals of the Keras Functional API has been completed, that should improve the reliability, stability, and performance of constructing Functional models.

  • Keras mixed precision API tf.keras.mixed_precision is no longer experimental and allows the use of 16-bit floating point formats during training, improving performance by up to 3x on GPUs and 60% on TPUs. Please see below for additional details.

  • TensorFlow Profiler now supports profiling MultiWorkerMirroredStrategy and tracing multiple workers using the sampling mode API.

  • TFLite Profiler for Android is available. See the detailed guide to learn more.

  • TensorFlow pip packages are now built with CUDA11 and cuDNN 8.0.2.

Breaking Changes

  • TF Core:

    • Certain float32 ops run in lower precision on Ampere based GPUs, including matmuls and convolutions, due to the use of TensorFloat-32. Specifically, inputs to such ops are rounded from 23 bits of precision to 10 bits of precision. This is unlikely to cause issues in practice for deep learning models. In some cases, TensorFloat-32 is also used for complex64 ops. TensorFloat-32 can be disabled by running tf.config.experimental.enable_tensor_float_32_execution(False).
    • The byte layout for string tensors across the C-API has been updated to match TF Core/C++; i.e., a contiguous array of tensorflow::tstring/TF_TStrings.
    • C-API functions TF_StringDecode, TF_StringEncode, and TF_StringEncodedSize are no longer relevant and have been removed; see core/platform/ctstring.h for string access/modification in C.
    • tensorflow.python, tensorflow.core and tensorflow.compiler modules are now hidden. These modules are not part of TensorFlow public API.
    • tf.raw_ops.Max and tf.raw_ops.Min no longer accept inputs of type tf.complex64 or tf.complex128, because the behavior of these ops is not well defined for complex types.
    • XLA:CPU and XLA:GPU devices are no longer registered by default. Use TF_XLA_FLAGS=--tf_xla_enable_xla_devices if you really need them, but this flag will eventually be removed in subsequent releases.
  • tf.keras:

    • The steps_per_execution argument in model.compile() is no longer experimental; if you were passing experimental_steps_per_execution, rename it to steps_per_execution in your code. This argument controls the number of batches to run during each tf.function call when calling model.fit(). Running multiple batches inside a single tf.function call can greatly improve performance on TPUs or small models with a large Python overhead.
    • A major refactoring of the internals of the Keras Functional API may affect code that is relying on certain internal details:
    • Code that uses isinstance(x, tf.Tensor) instead of tf.is_tensor when checking Keras symbolic inputs/outputs should switch to using tf.is_tensor.
    • Code that is overly dependent on the exact names attached to symbolic tensors (e.g. assumes there will be ":0" at the end of the inputs, treats names as unique identifiers instead of using tensor.ref(), etc.) may break.
    • Code that uses full path for get_concrete_function to trace Keras symbolic inputs directly should switch to building matching tf.TensorSpecs directly and tracing the TensorSpec objects.
    • Code that relies on the exact number and names of the op layers that TensorFlow operations were converted into may have changed.
    • Code that uses tf.map_fn/tf.cond/tf.while_loop/control flow as op layers and happens to work before TF 2.4. These will explicitly be unsupported now. Converting these ops to Functional API op layers was unreliable before TF 2.4, and prone to erroring incomprehensibly or being silently buggy.
    • Code that directly asserts on a Keras symbolic value in cases where ops like tf.rank used to return a static or symbolic value depending on if the input had a fully static shape or not. Now these ops always return symbolic values.
    • Code already susceptible to leaking tensors outside of graphs becomes slightly more likely to do so now.
    • Code that tries directly getting gradients with respect to symbolic Keras inputs/outputs. Use GradientTape on the actual Tensors passed to the already-constructed model instead.
    • Code that requires very tricky shape manipulation via converted op layers in order to work, where the Keras symbolic shape inference proves insufficient.
    • Code that tries manually walking a tf.keras.Model layer by layer and assumes layers only ever have one positional argument. This assumption doesn't hold true before TF 2.4 either, but is more likely to cause issues now.
    • Code that manually enters keras.backend.get_graph() before building a functional model is no longer needed.
    • Start enforcing input shape assumptions when calling Functional API Keras models. This may potentially break some users, in case there is a mismatch between the shape used when creating Input objects in a Functional model, and the shape of the data passed to that model. You can fix this mismatch by either calling the model with correctly-shaped data, or by relaxing Input shape assumptions (note that you can pass shapes with None entries for axes that are meant to be dynamic). You can also disable the input checking entirely by setting model.input_spec = None.
    • Several changes have been made to tf.keras.mixed_precision.experimental. Note that it is now recommended to use the non-experimental tf.keras.mixed_precision API.
    • AutoCastVariable.dtype now refers to the actual variable dtype, not the dtype it will be casted to.
    • When mixed precision is enabled, tf.keras.layers.Embedding now outputs a float16 or bfloat16 tensor instead of a float32 tensor.
    • The property tf.keras.mixed_precision.experimental.LossScaleOptimizer.loss_scale is now a tensor, not a LossScale object. This means to get a loss scale of a LossScaleOptimizer as a tensor, you must now call opt.loss_scaleinstead of opt.loss_scale().
    • The property should_cast_variables has been removed from tf.keras.mixed_precision.experimental.Policy
    • When passing a tf.mixed_precision.experimental.DynamicLossScale to tf.keras.mixed_precision.experimental.LossScaleOptimizer, the DynamicLossScale's multiplier must be 2.
    • When passing a tf.mixed_precision.experimental.DynamicLossScale to tf.keras.mixed_precision.experimental.LossScaleOptimizer, the weights of the DynanmicLossScale are copied into the LossScaleOptimizer instead of being reused. This means modifying the weights of the DynamicLossScale will no longer affect the weights of the LossScaleOptimizer, and vice versa.
    • The global policy can no longer be set to a non-floating point policy in tf.keras.mixed_precision.experimental.set_policy
    • In Layer.call, AutoCastVariables will no longer be casted within MirroredStrategy.run or ReplicaContext.merge_call. This is because a thread local variable is used to determine whether AutoCastVariables are casted, and those two functions run with a different thread. Note this only applies if one of these two functions is called within Layer.call; if one of those two functions calls Layer.call, AutoCastVariables will still be casted.
  • tf.data:

    • tf.data.experimental.service.DispatchServer now takes a config tuple instead of individual arguments. Usages should be updated to tf.data.experimental.service.DispatchServer(dispatcher_config).
    • tf.data.experimental.service.WorkerServer now takes a config tuple instead of individual arguments. Usages should be updated to tf.data.experimental.service.WorkerServer(worker_config).
  • tf.distribute:

    • Removes tf.distribute.Strategy.experimental_make_numpy_dataset. Please use tf.data.Dataset.from_tensor_slices instead.
    • Renames experimental_hints in tf.distribute.StrategyExtended.reduce_to, tf.distribute.StrategyExtended.batch_reduce_to, tf.distribute.ReplicaContext.all_reduce to options.
    • Renames tf.distribute.experimental.CollectiveHints to tf.distribute.experimental.CommunicationOptions.
    • Renames tf.distribute.experimental.CollectiveCommunication to tf.distribute.experimental.CommunicationImplementation.
    • Renames tf.distribute.Strategy.experimental_distribute_datasets_from_function to distribute_datasets_from_function as it is no longer experimental.
    • Removes tf.distribute.Strategy.experimental_run_v2 method, which was deprecated in TF 2.2.
  • tf.lite:

    • tf.quantization.quantize_and_dequantize_v2 has been introduced, which updates the gradient definition for quantization which is outside the range to be 0. To simulate the V1 the behavior of tf.quantization.quantize_and_dequantize(...) use tf.grad_pass_through(tf.quantization.quantize_and_dequantize_v2)(...).
  • Building TensorFlow:

    • Windows platform builds: TensorFlow on Windows under MSVC is now built with --copt=/experimental:preprocessor --host_copt=/experimental:preprocessor (see .bazelrc for more details). Builds including TensorFlow may fail with unexpected syntax errors if these flags are absent. See also this thread on SIG Build.

Known Caveats

  • tf.keras.mixed_precision
    • When using mixed precision, calling RMSprop.apply_gradients or Nadam.apply_gradients outside a tf.function does not work and will raise the AttributeError "Tensor.op is meaningless when eager execution is enabled". See this issue for details and a workaround.

Bug Fixes and Other Changes

TF Core:

  • Introduces experimental support for a new module named tf.experimental.numpy, which is a NumPy-compatible API for writing TF programs. This module provides class ndarray, which mimics the ndarray class in NumPy, and wraps an immutable tf.Tensor under the hood. A subset of NumPy functions (e.g. numpy.add) are provided. Their inter-operation with TF facilities is seamless in most cases. See tensorflow/python/ops/numpy_ops/README.md for details of what operations are supported and what are the differences from NumPy.
  • tf.types.experimental.TensorLike is a new Union type that can be used as type annotation for variables representing a Tensor or a value that can be converted to Tensor by tf.convert_to_tensor.
  • Calling ops with a python constants or numpy values is now consistent with tf.convert_to_tensor behavior. This avoids operations like tf.reshape truncating inputs such as from int64 to int32.
  • Adds tf.sparse.map_values to apply a function to the .values of SparseTensor arguments.
  • The Python bitwise operators for Tensor (__and__, __or__, __xor__ and __invert__ now support non-bool arguments and apply the corresponding bitwise ops. bool arguments continue to be supported and dispatch to logical ops. This brings them more in line with Python and NumPy behavior.
  • Adds tf.SparseTensor.with_values. This returns a new SparseTensor with the same sparsity pattern, but with new provided values. It is similar to the with_values function of RaggedTensor.
  • Adds StatelessCase op, and uses it if none of case branches has stateful ops.
  • Adds tf.config.experimental.get_memory_usage to return total memory usage of the device.
  • Adds gradients for RaggedTensorToVariant and RaggedTensorFromVariant.
  • Improve shape inference of nested function calls by supporting constant folding across Arg nodes which makes more static values available to shape inference functions.
  • tf.debugging:
    • tf.debugging.assert_shapes() now works on SparseTensors (Fixes #36268).
  • GPU
    • Adds Support for TensorFloat-32 on Ampere based GPUs.TensorFloat-32, or TF32 for short, is a math mode for NVIDIA Ampere based GPUs which causes certain float32 ops, such as matrix multiplications and convolutions, to run much faster on Ampere GPUs but with reduced precision. This reduced precision has not been found to effect convergence quality of deep learning models in practice. TensorFloat-32 is enabled by default, but can be disabled with tf.config.experimental.enable_tensor_float_32_execution.
  • tf.math:
    • Adds tf.math.erfcinv, the inverse to tf.math.erfc.
  • tf.nn:
    • tf.nn.max_pool2d now supports explicit padding.
  • tf.image:
    • Adds deterministic tf.image.stateless_random_* functions for each tf.image.random_* function. Added a new op stateless_sample_distorted_bounding_box which is a deterministic version of sample_distorted_bounding_box op. Given the same seed, these stateless functions/ops produce the same results independent of how many times the function is called, and independent of global seed settings.
    • Adds deterministic tf.image.resize backprop CUDA kernels for method=ResizeMethod.BILINEAR (the default method). Enable by setting the environment variable TF_DETERMINISTIC_OPS to "true" or "1".
  • tf.print:
    • Bug fix in tf.print() with OrderedDict where if an OrderedDict didn't have the keys sorted, the keys and values were not being printed in accordance with their correct mapping.
  • tf.train.Checkpoint:
    • Now accepts a root argument in the initialization, which generates a checkpoint with a root object. This allows users to create a Checkpoint object that is compatible with Keras model.save_weights() and model.load_weights. The checkpoint is also compatible with the checkpoint saved in the variables/ folder in the SavedModel.
    • When restoring, save_path can be a path to a SavedModel. The function will automatically find the checkpoint in the SavedModel.

tf.data:

  • Adds new tf.data.experimental.service.register_dataset and tf.data.experimental.service.from_dataset_id APIs to enable one process to register a dataset with the tf.data service, and another process to consume data from the dataset.
  • Adds support for dispatcher fault tolerance. To enable fault tolerance, configure a work_dir when running your dispatcher server and set dispatcher_fault_tolerance=True. The dispatcher will store its state to work_dir, so that on restart it can continue from its previous state after restart.
  • Adds support for sharing dataset graphs via shared filesystem instead of over RPC. This reduces load on the dispatcher, improving performance of distributing datasets. For this to work, the dispatcher's work_dir must be accessible from workers. If the worker fails to read from the work_dir, it falls back to using RPC for dataset graph transfer.
  • Adds support for a new "distributed_epoch" processing mode. This processing mode distributes a dataset across all tf.data workers, instead of having each worker process the full dataset. See the tf.data service docs to learn more.
  • Adds optional exclude_cols parameter to CsvDataset. This parameter is the complement of select_cols; at most one of these should be specified.
  • We have implemented an optimization which reorders data-discarding transformations such as take and shard to happen earlier in the dataset when it is safe to do so. The optimization can be disabled via the experimental_optimization.reorder_data_discarding_ops dataset option.
  • tf.data.Options were previously immutable and can now be overridden.
  • tf.data.Dataset.from_generator now supports Ragged and Sparse tensors with a new output_signature argument, which allows from_generator to produce any type describable by a tf.TypeSpec.
  • tf.data.experimental.AUTOTUNE is now available in the core API as tf.data.AUTOTUNE.

tf.distribute:

  • Introduces experimental support for asynchronous training of models via tf.distribute.experimental.ParameterServerStrategy:
    • Replaces the existing tf.distribute.experimental.ParameterServerStrategy symbol with a new class that is for parameter server training in TF2. Usage of the old symbol, usually with Estimator API, should be replaced with [tf.compat.v1.distribute.experimental.ParameterServerStrategy].
    • Added tf.distribute.experimental.coordinator.* namespace, including the main API ClusterCoordinator for coordinating the training cluster, the related data structure RemoteValue and PerWorkerValue.
  • MultiWorkerMirroredStrategy](https://www.tensorflow.org/api_docs/python/tf/distribute/MultiWorkerMirroredStrategy) is now a stable API and is no longer considered experimental. Some of the major improvements involve handling peer failure and many bug fixes. Please check out the detailed tutorial on Multi-worer training with Keras.
  • Adds tf.distribute.Strategy.gather and tf.distribute.ReplicaContext.all_gather APIs to support gathering dense distributed values.
  • Fixes various issues with saving a distributed model.

tf.keras:

  • Improvements from the Functional API refactoring:
    • Functional model construction does not need to maintain a global workspace graph, removing memory leaks especially when building many models or very large models.
    • Functional model construction should be ~8-10% faster on average.
    • Functional models can now contain non-symbolic values in their call inputs inside of the first positional argument.
    • Several classes of TF ops that were not reliably converted to Keras layers during functional API construction should now work, e.g.tf.image.ssim_multiscale
    • Error messages when Functional API construction goes wrong (and when ops cannot be converted to Keras layers automatically) should be clearer and easier to understand.
  • Optimizer.minimize can now accept a loss Tensor and a GradientTape as an alternative to accepting a callable loss.
  • Adds beta hyperparameter to FTRL optimizer classes (Keras and others) to match FTRL paper.
  • Optimizer.__init__ now accepts a gradient_aggregator to allow for customization of how gradients are aggregated across devices, as well as gradients_transformers to allow for custom gradient transformations (such as gradient clipping).
  • Improvements to Keras preprocessing layers:
    • TextVectorization can now accept a vocabulary list or file as an init arg.
    • Normalization can now accept mean and variance values as init args.
  • In Attention and AdditiveAttention layers, the call() method now accepts a return_attention_scores argument. When set to True, the layer returns the attention scores as an additional output argument.
  • Adds tf.metrics.log_cosh and tf.metrics.logcosh API entrypoints with the same implementation as their tf.losses equivalent.
  • For Keras model, the individual call of Model.evaluate uses no cached data for evaluation, while Model.fit uses cached data when validation_data arg is provided for better performance.
  • Adds a save_traces argument to model.save/ tf.keras.models.save_model which determines whether the SavedModel format stores the Keras model/layer call functions. The traced functions allow Keras to revive custom models and layers without the original class definition, but if this isn't required the tracing can be disabled with the added option.
  • The tf.keras.mixed_precision API is now non-experimental. The non-experimental API differs from the experimental API in several ways.
    • tf.keras.mixed_precision.Policy no longer takes in a tf.mixed_precision. experimental.LossScale in the constructor, and no longer has a LossScale associated with it. Instead, Model.compile will automatically wrap the optimizer with a LossScaleOptimizer using dynamic loss scaling if Policy.name is "mixed_float16".
    • tf.keras.mixed_precision.LossScaleOptimizer's constructor takes in different arguments. In particular, it no longer takes in a LossScale, and there is no longer a LossScale associated with the LossScaleOptimizer. Instead, LossScaleOptimizer directly implements fixed or dynamic loss scaling. See the documentation of tf.keras.mixed_precision.experimental.LossScaleOptimizer for details on the differences between the experimental LossScaleOptimizer and the new non-experimental LossScaleOptimizer.
    • tf.mixed_precision.experimental.LossScale and its subclasses are deprecated, as all of its functionality now exists within tf.keras.mixed_precision.LossScaleOptimizer

tf.lite:

  • TFLiteConverter:
    • Support optional flags inference_input_type and inference_output_type for full integer quantized models. This allows users to modify the model input and output type to integer types (tf.int8, tf.uint8) instead of defaulting to float type (tf.float32).
  • NNAPI
    • Adds NNAPI Delegation support for requantization use cases by converting the operation into a dequantize-quantize pair.
    • Removes deprecated Interpreter.setUseNNAPI(boolean) Java API. Use Interpreter.Options.setUseNNAPI instead.
    • Deprecates Interpreter::UseNNAPI(bool) C++ API. Use NnApiDelegate() and related delegate configuration methods directly.
    • Deprecates Interpreter::SetAllowFp16PrecisionForFp32(bool) C++ API. Prefer controlling this via delegate options, e.g. tflite::StatefulNnApiDelegate::Options::allow_fp16' orTfLiteGpuDelegateOptionsV2::is_precision_loss_allowed`.
  • GPU
    • GPU acceleration now supports quantized models by default
  • DynamicBuffer::AddJoinedString() will now add a separator if the first string to be joined is empty.
  • Adds support for cumulative sum (cumsum), both as builtin op and MLIR conversion.

TensorRT

  • Issues a warning when the session_config parameter for the TF1 converter is used or the rewrite_config_template field in the TF2 converter parameter object is used.

TPU Enhancements:

  • Adds support for the beta parameter of the FTRL optimizer for TPU embeddings. Users of other TensorFlow platforms can implement equivalent behavior by adjusting the l2 parameter.

XLA Support:

  • xla.experimental.compile is deprecated, use tf.function(experimental_compile=True) instead.
  • Adds tf.function.experimental_get_compiler_ir which returns compiler IR (currently 'hlo' and 'optimized_hlo') for given input for given function.

Security:

Other:

  • We have replaced uses of "whitelist" and "blacklist" with "allowlist" and "denylist" where possible. Please see this list for more context.
  • Adds tf.config.experimental.mlir_bridge_rollout which will help us rollout the new MLIR TPU bridge.
  • Adds tf.experimental.register_filesystem_plugin to load modular filesystem plugins from Python

Thanks to our Contributors

This release contains contributions from many people at Google as well as the following external contributors:

8bitmp3, aaa.jq, Abhineet Choudhary, Abolfazl Shahbazi, acxz, Adam Hillier, Adrian Garcia Badaracco, Ag Ramesh, ahmedsabie, Alan Anderson, Alexander Grund, Alexandre Lissy, Alexey Ivanov, Amedeo Cavallo, anencore94, Aniket Kumar Singh, Anthony Platanios, Ashwin Phadke, Balint Cristian, Basit Ayantunde, bbbboom, Ben Barsdell, Benjamin Chetioui, Benjamin Peterson, bhack, Bhanu Prakash Bandaru Venkata, Biagio Montaruli, Brent M. Spell, bubblebooy, bzhao, cfRod, Cheng Chen, Cheng(Kit) Chen, Chris Tessum, Christian, chuanqiw, codeadmin_peritiae, COTASPAR, CuiYifeng, danielknobe, danielyou0230, dannyfriar, daria, DarrenZhang01, Denisa Roberts, dependabot[bot], Deven Desai, Dmitry Volodin, Dmitry Zakharov, drebain, Duncan Riach, Eduard Feicho, Ehsan Toosi, Elena Zhelezina, emlaprise2358, Eugene Kuznetsov, Evaderan-Lab, Evgeniy Polyakov, Fausto Morales, Felix Johnny, fo40225, Frederic Bastien, Fredrik Knutsson, fsx950223, Gaurav Singh, Gauri1 Deshpande, George Grzegorz Pawelczak, gerbauz, Gianluca Baratti, Giorgio Arena, Gmc2, Guozhong Zhuang, Hannes Achleitner, Harirai, HarisWang, Harsh188, hedgehog91, Hemal Mamtora, Hideto Ueno, Hugh Ku, Ian Beauregard, Ilya Persky, jacco, Jakub Beránek, Jan Jongboom, Javier Montalt Tordera, Jens Elofsson, Jerry Shih, jerryyin, jgehw, Jinjing Zhou, jma, jmsmdy, Johan Nordström, John Poole, Jonah Kohn, Jonathan Dekhtiar, jpodivin, Jung Daun, Kai Katsumata, Kaixi Hou, Kamil Rakoczy, Kaustubh Maske Patil, Kazuaki Ishizaki, Kedar Sovani, Koan-Sin Tan, Koki Ibukuro, Krzysztof Laskowski, Kushagra Sharma, Kushan Ahmadian, Lakshay Tokas, Leicong Li, levinxo, Lukas Geiger, Maderator, Mahmoud Abuzaina, Mao Yunfei, Marius Brehler, markf, Martin Hwasser, Martin Kubovčík, Matt Conley, Matthias, mazharul, mdfaijul, Michael137, MichelBr, Mikhail Startsev, Milan Straka, Ml-0, Myung-Hyun Kim, Måns Nilsson, Nathan Luehr, ngc92, nikochiko, Niranjan Hasabnis, nyagato_00, Oceania2018, Oleg Guba, Ongun Kanat, OscarVanL, Patrik Laurell, Paul Tanger, Peter Sobot, Phil Pearl, PlusPlusUltra, Poedator, Prasad Nikam, Rahul-Kamat, Rajeshwar Reddy T, redwrasse, Rickard, Robert Szczepanski, Rohan Lekhwani, Sam Holt, Sami Kama, Samuel Holt, Sandeep Giri, sboshin, Sean Settle, settle, Sharada Shiddibhavi, Shawn Presser, ShengYang1, Shi,Guangyong, Shuxiang Gao, Sicong Li, Sidong-Wei, Srihari Humbarwadi, Srinivasan Narayanamoorthy, Steenu Johnson, Steven Clarkson, stjohnso98, Tamas Bela Feher, Tamas Nyiri, Tarandeep Singh, Teng Lu, Thibaut Goetghebuer-Planchon, Tim Bradley, Tomasz Strejczek, Tongzhou Wang, Torsten Rudolf, Trent Lo, Ty Mick, Tzu-Wei Sung, Varghese, Jojimon, Vignesh Kothapalli, Vishakha Agrawal, Vividha, Vladimir Menshakov, Vladimir Silyaev, VoVAllen, Võ Văn Nghĩa, wondertx, xiaohong1031, Xiaoming (Jason) Cui, Xinan Jiang, Yair Ehrenwald, Yasir Modak, Yasuhiro Matsumoto, Yimei Sun, Yiwen Li, Yixing, Yoav Ramon, Yong Tang, Yong Wu, yuanbopeng, Yunmo Koo, Zhangqiang, Zhou Peng, ZhuBaohe, zilinzhu, zmx

Release 2.3.1

Bug Fixes and Other Changes

Release 2.2.1

Bug Fixes and Other Changes

Release 2.1.2

Bug Fixes and Other Changes

Release 2.0.3

Bug Fixes and Other Changes

Release 1.15.4

Bug Fixes and Other Changes

Release 2.3.0

Major Features and Improvements

  • tf.data adds two new mechanisms to solve input pipeline bottlenecks and save resources:

    In addition checkout the detailed guide for analyzing input pipeline performance with TF Profiler.

  • tf.distribute.TPUStrategy is now a stable API and no longer considered experimental for TensorFlow. (earlier tf.distribute.experimental.TPUStrategy).

  • TF Profiler introduces two new tools: a memory profiler to visualize your model’s memory usage over time and a python tracer which allows you to trace python function calls in your model. Usability improvements include better diagnostic messages and profile options to customize the host and device trace verbosity level.

  • Introduces experimental support for Keras Preprocessing Layers API (tf.keras.layers.experimental.preprocessing.*) to handle data preprocessing operations, with support for composite tensor inputs. Please see below for additional details on these layers.

  • TFLite now properly supports dynamic shapes during conversion and inference. We’ve also added opt-in support on Android and iOS for XNNPACK, a highly optimized set of CPU kernels, as well as opt-in support for executing quantized models on the GPU.

  • Libtensorflow packages are available in GCS starting this release. We have also started to release a nightly version of these packages.

  • The experimental Python API tf.debugging.experimental.enable_dump_debug_info() now allows you to instrument a TensorFlow program and dump debugging information to a directory on the file system. The directory can be read and visualized by a new interactive dashboard in TensorBoard 2.3 called Debugger V2, which reveals the details of the TensorFlow program including graph structures, history of op executions at the Python (eager) and intra-graph levels, the runtime dtype, shape, and numerical composition of tensors, as well as their code locations.

Breaking Changes

  • Increases the minimum bazel version required to build TF to 3.1.0.
  • tf.data
    • Makes the following (breaking) changes to the tf.data.
    • C++ API: - IteratorBase::RestoreInternal, IteratorBase::SaveInternal, and DatasetBase::CheckExternalState become pure-virtual and subclasses are now expected to provide an implementation.
    • The deprecated DatasetBase::IsStateful method is removed in favor of DatasetBase::CheckExternalState.
    • Deprecated overrides of DatasetBase::MakeIterator and MakeIteratorFromInputElement are removed.
    • The signature of tensorflow::data::IteratorBase::SaveInternal and tensorflow::data::IteratorBase::SaveInput has been extended with SerializationContext argument to enable overriding the default policy for the handling external state during iterator checkpointing. This is not a backwards compatible change and all subclasses of IteratorBase need to be updated accordingly.
  • tf.keras
    • Add a new BackupAndRestore callback for handling distributed training failures & restarts. Please take a look at this tutorial for details on how to use the callback.
  • tf.image.extract_glimpse has been updated to correctly process the case where centered=False and normalized=False. This is a breaking change as the output is different from (incorrect) previous versions. Note this breaking change only impacts tf.image.extract_glimpse and tf.compat.v2.image.extract_glimpse API endpoints. The behavior of tf.compat.v1.image.extract_glimpse does not change. The behavior of existing C++ kernel ExtractGlimpse does not change either, so saved models using tf.raw_ops.ExtractGlimpse will not be impacted.

Known Caveats

  • tf.lite
    • Keras-based LSTM models must be converted with an explicit batch size in the input layer.

Bug Fixes and Other Changes

TF Core:

  • Set tf2_behavior to 1 to enable V2 for early loading cases.
  • Add execute_fn_for_device function to dynamically choose the implementation based on underlying device placement.
  • Eager:
    • Add reduce_logsumexp benchmark with experiment compile.
    • Give EagerTensors a meaningful __array__ implementation.
    • Add another version of defun matmul for performance analysis.
  • tf.function/AutoGraph:
    • AutoGraph now includes into TensorFlow loops any variables that are closed over by local functions. Previously, such variables were sometimes incorrectly ignored.
    • functions returned by the get_concrete_function method of tf.function objects can now be called with arguments consistent with the original arguments or type specs passed to get_concrete_function. This calling convention is now the preferred way to use concrete functions with nested values and composite tensors. Please check the guide for more details on concrete_ function.
    • Update tf.function's experimental_relax_shapes to handle composite tensors appropriately.
    • Optimize tf.function invocation, by removing redundant list converter.
    • tf.function will retrace when called with a different variable instead of simply using the dtype & shape.
    • Improve support for dynamically-sized TensorArray inside tf.function.
  • tf.math:
    • Narrow down argmin/argmax contract to always return the smallest index for ties.
    • tf.math.reduce_variance and tf.math.reduce_std return correct computation for complex types and no longer support integer types.
    • Add Bessel functions of order 0,1 to tf.math.special.
    • tf.divide now always returns a tensor to be consistent with documentation and other APIs.
  • tf.image:
    • Replaced tf.image.non_max_suppression_padded with a new implementation that supports batched inputs, which is considerably faster on TPUs and GPUs. Boxes with area=0 will be ignored. Existing usage with single inputs should still work as before.
  • tf.linalg
    • Add tf.linalg.banded_triangular_solve.
  • tf.random:
    • Add tf.random.stateless_parameterized_truncated_normal.
  • tf.ragged:
    • Add tf.ragged.cross and tf.ragged.cross_hashed operations.
  • tf.RaggedTensor:
    • RaggedTensor.to_tensor() now preserves static shape.
    • Add tf.strings.format() and tf.print() to support RaggedTensors.
  • tf.saved_model:
    • @tf.function from SavedModel no longer ignores args after a RaggedTensor when selecting the concrete function to run.
    • Fix save model issue for ops with a list of functions.
    • Add tf.saved_model.LoadOptions with experimental_io_device as arg with default value None to choose the I/O device for loading models and weights.
    • Update tf.saved_model.SaveOptions with experimental_io_device as arg with default value None to choose the I/O device for saving models and weights.
    • Mutable tables now restore checkpointed values when loaded from SavedModel.
    • The user object metadata field in the SavedModel proto has been deprecated as part of the updates to Keras SavedModel. Keras was the only consumer of this field prior to the update.
  • GPU
    • TF 2.3 includes PTX kernels only for compute capability 7.0 to reduce the TF pip binary size. Earlier releases included PTX for a variety of older compute capabilities.
    • Remove environmental variable TF_USE_CUDNN.
  • Others
    • Retain parent namescope for ops added inside tf.while_loop/tf.cond/tf.switch_case.
    • Update tf.vectorized_map to support vectorizing tf.while_loop and TensorList operations.
    • tf.custom_gradient can now be applied to functions that accept nested structures of tensors as inputs (instead of just a list of tensors). Note that Python structures such as tuples and lists now won't be treated as tensors, so if you still want them to be treated that way, you need to wrap them with tf.convert_to_tensor.
    • No lowering on gradient case op when input is DeviceIndex op.
    • Extend the ragged version of tf.gather to support batch_dims and axis args.
    • Update tf.map_fn to support RaggedTensors and SparseTensors.
    • Deprecate tf.group. It is not useful in eager mode.
    • Add CPU and GPU implementation of modified variation of FTRL/FTRLV2 that can triggerred by multiply_linear_by_lr allowing a learning rate of zero.

tf.data:

  • tf.data.experimental.dense_to_ragged_batch works correctly with tuples.
  • tf.data.experimental.dense_to_ragged_batch to output variable ragged rank.
  • tf.data.experimental.cardinality is now a method on tf.data.Dataset.
  • tf.data.Dataset now supports len(Dataset) when the cardinality is finite.

tf.distribute:

  • Expose experimental tf.distribute.DistributedDataset and tf.distribute.DistributedIterator to distribute input data when using tf.distribute to scale training on multiple devices.
  • Allow var.assign on MirroredVariables with aggregation=NONE in replica context. Previously this would raise an error. We now allow this because many users and library writers find using .assign in replica context to be more convenient, instead of having to use Strategy.extended.update which was the previous way of updating variables in this situation.
  • tf.distribute.experimental.MultiWorkerMirroredStrategy adds support for partial batches. Workers running out of data now continue to participate in the training with empty inputs, instead of raising an error. Learn more about partial batches here.
  • Improve the performance of reading metrics eagerly under tf.distribute.experimental.MultiWorkerMirroredStrategy.
  • Fix the issue that strategy.reduce() inside tf.function may raise exceptions when the values to reduce are from loops or if-clauses.
  • Fix the issue that tf.distribute.MirroredStrategy cannot be used together with tf.distribute.experimental.MultiWorkerMirroredStrategy.
  • Add a tf.distribute.cluster_resolver.TPUClusterResolver.connect API to simplify TPU initialization.
  • Add tf.distribute.Strategy.gather and tf.distribute.ReplicaContext.all_gather methods to gather and concatenate tf.distribute.DistributedValues across workers and devices.

tf.keras:

  • Introduces experimental preprocessing layers API (tf.keras.layers.experimental.preprocessing) to handle data preprocessing operations such as categorical feature encoding, text vectorization, data normalization, and data discretization (binning). The newly added layers provide a replacement for the legacy feature column API, and support composite tensor inputs.
  • Added categorical data processing layers:
    • IntegerLookup & StringLookup: build an index of categorical feature values
    • CategoryEncoding: turn integer-encoded categories into one-hot, multi-hot, or tf-idf encoded representations
    • CategoryCrossing: create new categorical features representing co-occurrences of previous categorical feature values
    • Hashing: the hashing trick, for large-vocabulary categorical features
    • Discretization: turn continuous numerical features into categorical features by binning their values
  • Improved image preprocessing layers: CenterCrop, Rescaling
  • Improved image augmentation layers: RandomCrop, RandomFlip, RandomTranslation, RandomRotation, RandomHeight, RandomWidth, RandomZoom, RandomContrast
  • Improved TextVectorization layer, which handles string tokenization, n-gram generation, and token encoding
    • The TextVectorization layer now accounts for the mask_token as part of the vocabulary size when output_mode='int'. This means that, if you have a max_tokens value of 5000, your output will have 5000 unique values (not 5001 as before).
    • Change the return value of TextVectorization.get_vocabulary() from byte to string. Users who previously were calling 'decode' on the output of this method should no longer need to do so.
  • Introduce new Keras dataset generation utilities :
    • image_dataset_from_directory is a utility based on tf.data.Dataset, meant to replace the legacy ImageDataGenerator. It takes you from a structured directory of images to a labeled dataset, in one function call. Note that it doesn't perform image data augmentation (which is meant to be done using preprocessing layers).
    • text_dataset_from_directory takes you from a structured directory of text files to a labeled dataset, in one function call.
    • timeseries_dataset_from_array is a tf.data.Dataset-based replacement of the legacy TimeseriesGenerator. It takes you from an array of timeseries data to a dataset of shifting windows with their targets.
  • Added experimental_steps_per_execution arg to model.compile to indicate the number of batches to run per tf.function call. This can speed up Keras Models on TPUs up to 3x.
  • Extends tf.keras.layers.Lambda layers to support multi-argument lambdas, and keyword arguments when calling the layer.
  • Functional models now get constructed if any tensor in a layer call's arguments/keyword arguments comes from a keras input. Previously the functional api would only work if all of the elements in the first argument to the layer came from a keras input.
  • Clean up BatchNormalization layer's trainable property to act like standard python state when it's used inside tf.functions (frozen at tracing time), instead of acting like a pseudo-variable whose updates kind of sometimes get reflected in already-traced tf.function traces.
  • Add the Conv1DTranspose layer.
  • Refine the semantics of SensitivitySpecificityBase derived metrics. See the updated API docstrings for tf.keras.metrics.SensitivityAtSpecificity and tf.keras.metrics.SpecificityAtSensitivty.

tf.lite:

  • Converter
    • Restored inference_input_type and inference_output_type flags in TF 2.x TFLiteConverter (backward compatible with TF 1.x) to support integer (tf.int8, tf.uint8) input and output types in post training full integer quantized models.
    • Added support for converting and resizing models with dynamic (placeholder) dimensions. Previously, there was only limited support for dynamic batch size, and even that did not guarantee that the model could be properly resized at runtime.
      • Enabled experimental support for a new quantization mode with 16-bit activations and 8-bit weights. See lite.OpsSet.EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8.
  • CPU
    • Fix an issue w/ dynamic weights and Conv2D on x86.
    • Add a runtime Android flag for enabling XNNPACK for optimized CPU performance.
    • Add a runtime iOS flag for enabling XNNPACK for optimized CPU performance.
    • Add a compiler flag to enable building a TFLite library that applies XNNPACK delegate automatically when the model has a fp32 operation.
  • GPU
    • Allow GPU acceleration starting with internal graph nodes
    • Experimental support for quantized models with the Android GPU delegate
    • Add GPU delegate whitelist.
    • Rename GPU whitelist -> compatibility (list).
    • Improve GPU compatibility list entries from crash reports.
  • NNAPI
    • Set default value for StatefulNnApiDelegate::Options::max_number_delegated_partitions to 3.
    • Add capability to disable NNAPI CPU and check NNAPI Errno.
    • Fix crashes when using NNAPI with target accelerator specified with model containing Conv2d or FullyConnected or LSTM nodes with quantized weights.
    • Fix ANEURALNETWORKS_BAD_DATA execution failures with sum/max/min/reduce operations with scalar inputs.
  • Hexagon
    • TFLite Hexagon Delegate out of experimental.
    • Experimental int8 support for most hexagon ops.
    • Experimental per-channel quant support for conv in Hexagon delegate.
    • Support dynamic batch size in C++ API.
  • CoreML
    • Opensource CoreML delegate
  • Misc
    • Enable building Android TFLite targets on Windows
    • Add support for BatchMatMul.
    • Add support for half_pixel_centers with ResizeNearestNeighbor.
    • Add 3D support for BatchToSpaceND.
    • Add 5D support for BroadcastSub, Maximum, Minimum, Transpose and BroadcastDiv.
    • Rename kTfLiteActRelu1 to kTfLiteActReluN1To1.
    • Enable flex delegate on tensorflow.lite.Interpreter Python package.
    • Add Buckettize, SparseCross and BoostedTreesBucketize to the flex whitelist.
    • Add support for selective registration of flex ops.
    • Add missing kernels for flex delegate whitelisted ops.
    • Fix issue when using direct ByteBuffer inputs with graphs that have dynamic shapes.
    • Fix error checking supported operations in a model containing HardSwish.

Packaging Support

  • Added tf.sysconfig.get_build_info(). Returns a dict that describes the build environment of the currently installed TensorFlow package, e.g. the NVIDIA CUDA and NVIDIA CuDNN versions used when TensorFlow was built.

Profiler

  • Fix a subtle use-after-free issue in XStatVisitor::RefValue().

TPU Enhancements

  • Adds 3D mesh support in TPU configurations ops.
  • Added TPU code for FTRL with multiply_linear_by_lr.
  • Silently adds a new file system registry at gstpu.
  • Support restartType in cloud tpu client.
  • Depend on a specific version of google-api-python-client.
  • Fixes apiclient import.

Tracing and Debugging

  • Add a TFE_Py_Execute traceme.

XLA Support

  • Implement stable argmin and argmax

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

902449@58880@bigcat_chen@ASIC, Abdul Baseer Khan, Abhineet Choudhary, Abolfazl Shahbazi, Adam Hillier, ag.ramesh, Agoniii, Ajay P, Alex Hoffman, Alexander Bayandin, Alexander Grund, Alexandre Abadie, Alexey Rogachevskiy, amoitra, Andrew Stevens, Angus-Luo, Anshuman Tripathy, Anush Elangovan, Artem Mavrin, Ashutosh Hathidara, autoih, Ayushman Kumar, ayushmankumar7, Bairen Yi, Bas Aarts, Bastian Eichenberger, Ben Barsdell, bhack, Bharat Raghunathan, Biagio Montaruli, Bigcat-Himax, blueyi, Bryan Cutler, Byambaa, Carlos Hernandez-Vaquero, Chen Lei, Chris Knorowski, Christian Clauss, chuanqiw, CuiYifeng, Daniel Situnayake, Daria Zhuravleva, Dayananda-V, Deven Desai, Devi Sandeep Endluri, Dmitry Zakharov, Dominic Jack, Duncan Riach, Edgar Liberis, Ehsan Toosi, ekuznetsov139, Elena Zhelezina, Eugene Kuznetsov, Eugene Mikhantiev, Evgenii Zheltonozhskii, Fabio Di Domenico, Fausto Morales, Fei Sun, feihugis, Felix E. Klee, flyingcat, Frederic Bastien, Fredrik Knutsson, frreiss, fsx950223, ganler, Gaurav Singh, Georgios Pinitas, Gian Marco Iodice, Giorgio Arena, Giuseppe Rossini, Gregory Keith, Guozhong Zhuang, gurushantj, Hahn Anselm, Harald Husum, Harjyot Bagga, Hristo Vrigazov, Ilya Persky, Ir1d, Itamar Turner-Trauring, jacco, Jake Tae, Janosh Riebesell, Jason Zaman, jayanth, Jeff Daily, Jens Elofsson, Jinzhe Zeng, JLZ, Jonas Skog, Jonathan Dekhtiar, Josh Meyer, Joshua Chia, Judd, justkw, Kaixi Hou, Kam D Kasravi, Kamil Rakoczy, Karol Gugala, Kayou, Kazuaki Ishizaki, Keith Smiley, Khaled Besrour, Kilaru Yasaswi Sri Chandra Gandhi, Kim, Young Soo, Kristian Hartikainen, Kwabena W. Agyeman, Leslie-Fang, Leslie-Fang-Intel, Li, Guizi, Lukas Geiger, Lutz Roeder, M\U00E5Ns Nilsson, Mahmoud Abuzaina, Manish, Marcel Koester, Marcin Sielski, marload, Martin Jul, Matt Conley, mdfaijul, Meng, Peng, Meteorix, Michael Käufl, Michael137, Milan Straka, Mitchell Vitez, Ml-0, Mokke Meguru, Mshr-H, nammbash, Nathan Luehr, naumkin, Neeraj Bhadani, ngc92, Nick Morgan, nihui, Niranjan Hasabnis, Niranjan Yadla, Nishidha Panpaliya, Oceania2018, oclyke, Ouyang Jin, OverLordGoldDragon, Owen Lyke, Patrick Hemmer, Paul Andrey, Peng Sun, periannath, Phil Pearl, Prashant Dandriyal, Prashant Kumar, Rahul Huilgol, Rajan Singh, Rajeshwar Reddy T, rangjiaheng, Rishit Dagli, Rohan Reddy, rpalakkal, rposts, Ruan Kunliang, Rushabh Vasani, Ryohei Ikegami, Semun Lee, Seo-Inyoung, Sergey Mironov, Sharada Shiddibhavi, ShengYang1, Shraiysh Vaishay, Shunya Ueta, shwetaoj, Siyavash Najafzade, Srinivasan Narayanamoorthy, Stephan Uphoff, storypku, sunchenggen, sunway513, Sven-Hendrik Haase, Swapnil Parekh, Tamas Bela Feher, Teng Lu, tigertang, tomas, Tomohiro Ubukata, tongxuan.ltx, Tony Tonev, Tzu-Wei Huang, Téo Bouvard, Uday Bondhugula, Vaibhav Jade, Vijay Tadikamalla, Vikram Dattu, Vincent Abriou, Vishnuvardhan Janapati, Vo Van Nghia, VoVAllen, Will Battel, William D. Irons, wyzhao, Xiaoming (Jason) Cui, Xiaoquan Kong, Xinan Jiang, xutianming, Yair Ehrenwald, Yasir Modak, Yasuhiro Matsumoto, Yixing Fu, Yong Tang, Yuan Tang, zhaozheng09, Zilin Zhu, zilinzhu, 张志豪

Release 2.1.1

Bug Fixes and Other Changes

Release 2.0.2

Bug Fixes and Other Changes

Release 1.15.3

Bug Fixes and Other Changes

Release 2.2.0

TensorFlow 2.2 discontinues support for Python 2, previously announced as following Python 2's EOL on January 1, 2020.

Coinciding with this change, new releases of TensorFlow's Docker images provide Python 3 exclusively. Because all images now use Python 3, Docker tags containing -py3 will no longer be provided and existing -py3 tags like latest-py3 will not be updated.

Major Features and Improvements

  • Replaced the scalar type for string tensors from std::string to tensorflow::tstring which is now ABI stable.

  • A new Profiler for TF 2 for CPU/GPU/TPU. It offers both device and host performance analysis, including input pipeline and TF Ops. Optimization advisory is provided whenever possible. Please see this tutorial and guide for usage guidelines.

  • Export C++ functions to Python using pybind11 as opposed to SWIG as a part of our deprecation of swig efforts.

  • tf.distribute:

    • Support added for global sync BatchNormalization by using the newly added tf.keras.layers.experimental.SyncBatchNormalization layer. This layer will sync BatchNormalization statistics every step across all replicas taking part in sync training.
    • Performance improvements for GPU multi-worker distributed training using tf.distribute.experimental.MultiWorkerMirroredStrategy
    • Update NVIDIA NCCL to 2.5.7-1 for better performance and performance tuning. Please see nccl developer guide for more information on this.
    • Support gradient allreduce in float16. See this example usage.
    • Experimental support of all reduce gradient packing to allow overlapping gradient aggregation with backward path computation.
    • Deprecated experimental_run_v2 method for distribution strategies and renamed the method run as it is no longer experimental.
    • Add CompositeTensor support for DistributedIterators. This should help prevent unnecessary function retracing and memory leaks.
  • tf.keras:

    • Model.fit major improvements:
      • You can now use custom training logic with Model.fit by overriding Model.train_step.
      • Easily write state-of-the-art training loops without worrying about all of the features Model.fit handles for you (distribution strategies, callbacks, data formats, looping logic, etc)
      • See the default Model.train_step for an example of what this function should look like. Same applies for validation and inference via Model.test_step and Model.predict_step.
      • SavedModel uses its own Model._saved_model_inputs_spec attr now instead of relying on Model.inputs and Model.input_names, which are no longer set for subclass Models. This attr is set in eager, tf.function, and graph modes. This gets rid of the need for users to manually call Model._set_inputs when using Custom Training Loops(CTLs).
      • Dynamic shapes are supported for generators by calling the Model on the first batch we "peek" from the generator. This used to happen implicitly in Model._standardize_user_data. Long-term, a solution where the DataAdapter doesn't need to call the Model is probably preferable.
    • The SavedModel format now supports all Keras built-in layers (including metrics, preprocessing layers, and stateful RNN layers)
    • Update Keras batch normalization layer to use the running mean and average computation in the fused_batch_norm. You should see significant performance improvements when using fused_batch_norm in Eager mode.
  • tf.lite:

    • Enable TFLite experimental new converter by default.
  • XLA

    • XLA now builds and works on windows. All prebuilt packages come with XLA available.
    • XLA can be enabled for a tf.function with “compile or throw exception” semantics on CPU and GPU.

Breaking Changes

  • tf.keras:
    • In tf.keras.applications the name of the "top" layer has been standardized to "predictions". This is only a problem if your code relies on the exact name of the layer.
    • Huber loss function has been updated to be consistent with other Keras losses. It now computes mean over the last axis of per-sample losses before applying the reduction function.
  • AutoGraph no longer converts functions passed to tf.py_function, tf.py_func and tf.numpy_function.
  • Deprecating XLA_CPU and XLA_GPU devices with this release.
  • Increasing the minimum bazel version to build TF to 2.0.0 to use Bazel's cc_experimental_shared_library.
  • Keras compile/fit behavior for functional and subclassed models have been unified. Model properties such as metrics, metrics_names will now be available only after training/evaluating the model on actual data for functional models. metrics will now include model loss and output losses.loss_functions property has been removed from the model. This was an undocumented property that was accidentally public and has now been removed.

Known Caveats

  • The current TensorFlow release now requires gast version 0.3.3.

Bug Fixes and Other Changes

  • tf.data:
    • Removed autotune_algorithm from experimental optimization options.
  • TF Core:
    • tf.constant always creates CPU tensors irrespective of the current device context.
    • Eager TensorHandles maintain a list of mirrors for any copies to local or remote devices. This avoids any redundant copies due to op execution.
    • For tf.Tensor & tf.Variable, .experimental_ref() is no longer experimental and is available as simply .ref().
    • pfor/vectorized_map: Added support for vectorizing 56 more ops. Vectorizing tf.cond is also supported now.
    • Set as much partial shape as we can infer statically within the gradient impl of the gather op.
    • Gradient of tf.while_loop emits StatelessWhile op if cond and body functions are stateless. This allows multiple gradients while ops to run in parallel under distribution strategy.
    • Speed up GradientTape in eager mode by auto-generating list of op inputs/outputs which are unused and hence not cached for gradient functions.
    • Support back_prop=False in while_v2 but mark it as deprecated.
    • Improve error message when attempting to use None in data-dependent control flow.
    • Add RaggedTensor.numpy().
    • Update RaggedTensor.__getitem__ to preserve uniform dimensions & allow indexing into uniform dimensions.
    • Update tf.expand_dims to always insert the new dimension as a non-ragged dimension.
    • Update tf.embedding_lookup to use partition_strategy and max_norm when ids is ragged.
    • Allow batch_dims==rank(indices) in tf.gather.
    • Add support for bfloat16 in tf.print.
  • tf.distribute:
    • Support embedding_column with variable-length input features for MultiWorkerMirroredStrategy.
  • tf.keras:
    • Added experimental_aggregate_gradients argument to tf.keras.optimizer.Optimizer.apply_gradients. This allows custom gradient aggregation and processing aggregated gradients in custom training loop.
    • Allow pathlib.Path paths for loading models via Keras API.
  • tf.function/AutoGraph:
    • AutoGraph is now available in ReplicaContext.merge_call, Strategy.extended.update and Strategy.extended.update_non_slot.
    • Experimental support for shape invariants has been enabled in tf.function. See the API docs for tf.autograph.experimental.set_loop_options for additional info.
    • AutoGraph error messages now exclude frames corresponding to APIs internal to AutoGraph.
    • Improve shape inference for tf.function input arguments to unlock more Grappler optimizations in TensorFlow 2.x.
    • Improve automatic control dependency management of resources by allowing resource reads to occur in parallel and synchronizing only on writes.
    • Fix execution order of multiple stateful calls to experimental_run_v2 in tf.function.
    • You can now iterate over RaggedTensors using a for loop inside tf.function.
  • tf.lite:
    • Migrated the tf.lite C inference API out of experimental into lite/c.
    • Add an option to disallow NNAPI CPU / partial acceleration on Android 10
    • TFLite Android AARs now include the C headers and APIs are required to use TFLite from native code.
    • Refactors the delegate and delegate kernel sources to allow usage in the linter.
    • Limit delegated ops to actually supported ones if a device name is specified or NNAPI CPU Fallback is disabled.
    • TFLite now supports tf.math.reciprocal1 op by lowering to tf.div op.
    • TFLite's unpack op now supports boolean tensor inputs.
    • Microcontroller and embedded code moved from experimental to main TensorFlow Lite folder
    • Check for large TFLite tensors.
    • Fix GPU delegate crash with C++17.
    • Add 5D support to TFLite strided_slice.
    • Fix error in delegation of DEPTH_TO_SPACE to NNAPI causing op not to be accelerated.
    • Fix segmentation fault when running a model with LSTM nodes using NNAPI Delegate
    • Fix NNAPI delegate failure when an operand for Maximum/Minimum operation is a scalar.
    • Fix NNAPI delegate failure when Axis input for reduce operation is a scalar.
    • Expose option to limit the number of partitions that will be delegated to NNAPI.
    • If a target accelerator is specified, use its feature level to determine operations to delegate instead of SDK version.
  • tf.random:
    • Various random number generation improvements:
    • Add a fast path for default random_uniform
    • random_seed documentation improvement.
    • RandomBinomial broadcasts and appends the sample shape to the left rather than the right.
    • Added tf.random.stateless_binomial, tf.random.stateless_gamma, tf.random.stateless_poisson
    • tf.random.stateless_uniform now supports unbounded sampling of int types.
  • Math and Linear Algebra:
    • Add tf.linalg.LinearOperatorTridiag.
    • Add LinearOperatorBlockLowerTriangular
    • Add broadcasting support to tf.linalg.triangular_solve#26204, tf.math.invert_permutation.
    • Add tf.math.sobol_sample op.
    • Add tf.math.xlog1py.
    • Add tf.math.special.{dawsn,expi,fresnel_cos,fresnel_sin,spence}.
    • Add a Modified Discrete Cosine Transform (MDCT) and its inverse to tf.signal.
  • TPU Enhancements:
    • Refactor TpuClusterResolver to move shared logic to a separate pip package.
    • Support configuring TPU software version from cloud tpu client.
    • Allowed TPU embedding weight decay factor to be multiplied by learning rate.
  • XLA Support:
    • Add standalone XLA AOT runtime target + relevant .cc sources to pip package.
    • Add check for memory alignment to MemoryAllocation::MemoryAllocation() on 32-bit ARM. This ensures a deterministic early exit instead of a hard to debug bus error later.
    • saved_model_cli aot_compile_cpu allows you to compile saved models to XLA header+object files and include them in your C++ programs.
    • Enable Igamma, Igammac for XLA.
  • Deterministic Op Functionality:
    • XLA reduction emitter is deterministic when the environment variable TF_DETERMINISTIC_OPS is set to "true" or "1". This extends deterministic tf.nn.bias_add back-prop functionality (and therefore also deterministic back-prop of bias-addition in Keras layers) to include when XLA JIT compilation is enabled.
    • Fix problem, when running on a CUDA GPU and when either environment variable TF_DETERMINISTIC_OPS or environment variable TF_CUDNN_DETERMINISTIC is set to "true" or "1", in which some layer configurations led to an exception with the message "No algorithm worked!"
  • Tracing and Debugging:
    • Add source, destination name to _send traceme to allow easier debugging.
    • Add traceme event to fastpathexecute.
  • Other:
    • Fix an issue with AUC.reset_states for multi-label AUC #35852
    • Fix the TF upgrade script to not delete files when there is a parsing error and the output mode is in-place.
    • Move tensorflow/core:framework/*_pyclif rules to tensorflow/core/framework:*_pyclif.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

372046933, 8bitmp3, aaronhma, Abin Shahab, Aditya Patwardhan, Agoniii, Ahti Kitsik, Alan Yee, Albin Joy, Alex Hoffman, Alexander Grund, Alexandre E. Eichenberger, Amit Kumar Jaiswal, amoitra, Andrew Anderson, Angus-Luo, Anthony Barbier, Anton Kachatkou, Anuj Rawat, archis, Arpan-Dhatt, Arvind Sundararajan, Ashutosh Hathidara, autoih, Bairen Yi, Balint Cristian, Bas Aarts, BashirSbaiti, Basit Ayantunde, Ben Barsdell, Benjamin Gaillard, boron, Brett Koonce, Bryan Cutler, Christian Goll, Christian Sachs, Clayne Robison, comet, Daniel Falbel, Daria Zhuravleva, darsh8200, David Truby, Dayananda-V, deepakm, Denis Khalikov, Devansh Singh, Dheeraj R Reddy, Diederik Van Liere, Diego Caballero, Dominic Jack, dothinking, Douman, Drake Gens, Duncan Riach, Ehsan Toosi, ekuznetsov139, Elena Zhelezina, elzino, Ending2015a, Eric Schweitz, Erik Zettel, Ethan Saadia, Eugene Kuznetsov, Evgeniy Zheltonozhskiy, Ewout Ter Hoeven, exfalso, FAIJUL, Fangjun Kuang, Fei Hu, Frank Laub, Frederic Bastien, Fredrik Knutsson, frreiss, Frédéric Rechtenstein, fsx950223, Gaurav Singh, gbaned, George Grzegorz Pawelczak, George Sterpu, Gian Marco Iodice, Giorgio Arena, Hans Gaiser, Hans Pabst, Haoyu Wu, Harry Slatyer, hsahovic, Hugo, Hugo Sjöberg, IrinaM21, jacco, Jake Tae, Jean-Denis Lesage, Jean-Michel Gorius, Jeff Daily, Jens Elofsson, Jerry Shih, jerryyin, Jin Mingjian, Jinjing Zhou, JKIsaacLee, jojimonv, Jonathan Dekhtiar, Jose Ignacio Gomez, Joseph-Rance, Judd, Julian Gross, Kaixi Hou, Kaustubh Maske Patil, Keunwoo Choi, Kevin Hanselman, Khor Chean Wei, Kilaru Yasaswi Sri Chandra Gandhi, Koan-Sin Tan, Koki Ibukuro, Kristian Holsheimer, kurileo, Lakshay Tokas, Lee Netherton, leike666666, Leslie-Fang-Intel, Li, Guizi, LIUJIAN435, Lukas Geiger, Lyo Nguyen, madisetti, Maher Jendoubi, Mahmoud Abuzaina, Manuel Freiberger, Marcel Koester, Marco Jacopo Ferrarotti, Markus Franke, marload, Mbah-Javis, mbhuiyan, Meng Zhang, Michael Liao, MichaelKonobeev, Michal Tarnowski, Milan Straka, minoring, Mohamed Nour Abouelseoud, MoussaMM, Mrinal Jain, mrTsjolder, Måns Nilsson, Namrata Bhave, Nicholas Gao, Niels Ole Salscheider, nikochiko, Niranjan Hasabnis, Nishidha Panpaliya, nmostafa, Noah Trenaman, nuka137, Officium, Owen L - Sfe, Pallavi G, Paul Andrey, Peng Sun, Peng Wu, Phil Pearl, PhilipMay, pingsutw, Pooya Davoodi, PragmaTwice, pshiko, Qwerty71, R Gomathi, Rahul Huilgol, Richard Xiao, Rick Wierenga, Roberto Rosmaninho, ruchit2801, Rushabh Vasani, Sami, Sana Damani, Sarvesh Dubey, Sasan Jafarnejad, Sergii Khomenko, Shane Smiskol, Shaochen Shi, sharkdtu, Shawn Presser, ShengYang1, Shreyash Patodia, Shyam Sundar Dhanabalan, Siju Samuel, Somyajit Chakraborty Sam, Srihari Humbarwadi, srinivasan.narayanamoorthy, Srishti Yadav, Steph-En-M, Stephan Uphoff, Stephen Mugisha, SumanSudhir, Taehun Kim, Tamas Bela Feher, TengLu, Tetragramm, Thierry Herrmann, Tian Jin, tigertang, Tom Carchrae, Tom Forbes, Trent Lo, Victor Peng, vijayphoenix, Vincent Abriou, Vishal Bhola, Vishnuvardhan Janapati, vladbataev, VoVAllen, Wallyss Lima, Wen-Heng (Jack) Chung, wenxizhu, William D. Irons, William Zhang, Xiaoming (Jason) Cui, Xiaoquan Kong, Xinan Jiang, Yasir Modak, Yasuhiro Matsumoto, Yaxun (Sam) Liu, Yong Tang, Ytyt-Yt, yuan, Yuan Mingshuai, Yuan Tang, Yuki Ueda, Yusup, zhangshijin, zhuwenxi

Release 2.0.1

Bug Fixes and Other Changes

Release 1.15.2

Bug Fixes and Other Changes

Release 2.1.0

TensorFlow 2.1 will be the last TF release supporting Python 2. Python 2 support officially ends an January 1, 2020. As announced earlier, TensorFlow will also stop supporting Python 2 starting January 1, 2020, and no more releases are expected in 2019.

Major Features and Improvements

  • The tensorflow pip package now includes GPU support by default (same as tensorflow-gpu) for both Linux and Windows. This runs on machines with and without NVIDIA GPUs. tensorflow-gpu is still available, and CPU-only packages can be downloaded at tensorflow-cpu for users who are concerned about package size.
  • Windows users: Officially-released tensorflow Pip packages are now built with Visual Studio 2019 version 16.4 in order to take advantage of the new /d2ReducedOptimizeHugeFunctions compiler flag. To use these new packages, you must install "Microsoft Visual C++ Redistributable for Visual Studio 2015, 2017 and 2019", available from Microsoft's website here.
    • This does not change the minimum required version for building TensorFlow from source on Windows, but builds enabling EIGEN_STRONG_INLINE can take over 48 hours to compile without this flag. Refer to configure.py for more information about EIGEN_STRONG_INLINE and /d2ReducedOptimizeHugeFunctions.
    • If either of the required DLLs, msvcp140.dll (old) or msvcp140_1.dll (new), are missing on your machine, import tensorflow will print a warning message.
  • The tensorflow pip package is built with CUDA 10.1 and cuDNN 7.6.
  • tf.keras
    • Experimental support for mixed precision is available on GPUs and Cloud TPUs. See usage guide.
    • Introduced the TextVectorization layer, which takes as input raw strings and takes care of text standardization, tokenization, n-gram generation, and vocabulary indexing. See this end-to-end text classification example.
    • Keras .compile .fit .evaluate and .predict are allowed to be outside of the DistributionStrategy scope, as long as the model was constructed inside of a scope.
    • Experimental support for Keras .compile, .fit, .evaluate, and .predict is available for Cloud TPUs, Cloud TPU, for all types of Keras models (sequential, functional and subclassing models).
    • Automatic outside compilation is now enabled for Cloud TPUs. This allows tf.summary to be used more conveniently with Cloud TPUs.
    • Dynamic batch sizes with DistributionStrategy and Keras are supported on Cloud TPUs.
    • Support for .fit, .evaluate, .predict on TPU using numpy data, in addition to tf.data.Dataset.
    • Keras reference implementations for many popular models are available in the TensorFlow Model Garden.
  • tf.data
    • Changes rebatching for tf.data datasets + DistributionStrategy for better performance. Note that the dataset also behaves slightly differently, in that the rebatched dataset cardinality will always be a multiple of the number of replicas.
    • tf.data.Dataset now supports automatic data distribution and sharding in distributed environments, including on TPU pods.
    • Distribution policies for tf.data.Dataset can now be tuned with 1. tf.data.experimental.AutoShardPolicy(OFF, AUTO, FILE, DATA) 2. tf.data.experimental.ExternalStatePolicy(WARN, IGNORE, FAIL)
  • tf.debugging
    • Add tf.debugging.enable_check_numerics() and tf.debugging.disable_check_numerics() to help debugging the root causes of issues involving infinities and NaNs.
  • tf.distribute
    • Custom training loop support on TPUs and TPU pods is available through strategy.experimental_distribute_dataset, strategy.experimental_distribute_datasets_from_function, strategy.experimental_run_v2, strategy.reduce.
    • Support for a global distribution strategy through tf.distribute.experimental_set_strategy(), in addition to strategy.scope().
  • TensorRT
    • TensorRT 6.0 is now supported and enabled by default. This adds support for more TensorFlow ops including Conv3D, Conv3DBackpropInputV2, AvgPool3D, MaxPool3D, ResizeBilinear, and ResizeNearestNeighbor. In addition, the TensorFlow-TensorRT python conversion API is exported as tf.experimental.tensorrt.Converter.
  • Environment variable TF_DETERMINISTIC_OPS has been added. When set to "true" or "1", this environment variable makes tf.nn.bias_add operate deterministically (i.e. reproducibly), but currently only when XLA JIT compilation is not enabled. Setting TF_DETERMINISTIC_OPS to "true" or "1" also makes cuDNN convolution and max-pooling operate deterministically. This makes Keras Conv*D and MaxPool*D layers operate deterministically in both the forward and backward directions when running on a CUDA-enabled GPU.

Breaking Changes

  • Deletes Operation.traceback_with_start_lines for which we know of no usages.
  • Removed id from tf.Tensor.__repr__() as id is not useful other than internal debugging.
  • Some tf.assert_* methods now raise assertions at operation creation time if the input tensors' values are known at that time, not during the session.run(). This only changes behavior when the graph execution would have resulted in an error. When this happens, a noop is returned and the input tensors are marked non-feedable. In other words, if they are used as keys in feed_dict argument to session.run(), an error will be raised. Also, because some assert ops don't make it into the graph, the graph structure changes. A different graph can result in different per-op random seeds when they are not given explicitly (most often).
  • The following APIs are not longer experimental: tf.config.list_logical_devices, tf.config.list_physical_devices, tf.config.get_visible_devices, tf.config.set_visible_devices, tf.config.get_logical_device_configuration, tf.config.set_logical_device_configuration.
  • tf.config.experimentalVirtualDeviceConfiguration has been renamed to tf.config.LogicalDeviceConfiguration.
  • tf.config.experimental_list_devices has been removed, please use tf.config.list_logical_devices.

Bug Fixes and Other Changes

  • tf.data
    • Fixes concurrency issue with tf.data.experimental.parallel_interleave with sloppy=True.
    • Add tf.data.experimental.dense_to_ragged_batch().
    • Extend tf.data parsing ops to support RaggedTensors.
  • tf.distribute
    • Fix issue where GRU would crash or give incorrect output when a tf.distribute.Strategy was used.
  • tf.estimator
    • Added option in tf.estimator.CheckpointSaverHook to not save the GraphDef.
    • Moving the checkpoint reader from swig to pybind11.
  • tf.keras
    • Export depthwise_conv2d in tf.keras.backend.
    • In Keras Layers and Models, Variables in trainable_weights, non_trainable_weights, and weights are explicitly deduplicated.
    • Keras model.load_weights now accepts skip_mismatch as an argument. This was available in external Keras, and has now been copied over to tf.keras.
    • Fix the input shape caching behavior of Keras convolutional layers.
    • Model.fit_generator, Model.evaluate_generator, Model.predict_generator, Model.train_on_batch, Model.test_on_batch, and Model.predict_on_batch methods now respect the run_eagerly property, and will correctly run using tf.function by default. Note that Model.fit_generator, Model.evaluate_generator, and Model.predict_generator are deprecated endpoints. They are subsumed by Model.fit, Model.evaluate, and Model.predict which now support generators and Sequences.
  • tf.lite
    • Legalization for NMS ops in TFLite.
    • add narrow_range and axis to quantize_v2 and dequantize ops.
    • Added support for FusedBatchNormV3 in converter.
    • Add an errno-like field to NNAPI delegate for detecting NNAPI errors for fallback behaviour.
    • Refactors NNAPI Delegate to support detailed reason why an operation is not accelerated.
    • Converts hardswish subgraphs into atomic ops.
  • Other
    • Critical stability updates for TPUs, especially in cases where the XLA compiler produces compilation errors.
    • TPUs can now be re-initialized multiple times, using tf.tpu.experimental.initialize_tpu_system.
    • Add RaggedTensor.merge_dims().
    • Added new uniform_row_length row-partitioning tensor to RaggedTensor.
    • Add shape arg to RaggedTensor.to_tensor; Improve speed of RaggedTensor.to_tensor.
    • tf.io.parse_sequence_example and tf.io.parse_single_sequence_example now support ragged features.
    • Fix while_v2 with variables in custom gradient.
    • Support taking gradients of V2 tf.cond and tf.while_loop using LookupTable.
    • Fix bug where vectorized_map failed on inputs with unknown static shape.
    • Add preliminary support for sparse CSR matrices.
    • Tensor equality with None now behaves as expected.
    • Make calls to tf.function(f)(), tf.function(f).get_concrete_function and tf.function(f).get_initialization_function thread-safe.
    • Extend tf.identity to work with CompositeTensors (such as SparseTensor)
    • Added more dtypes and zero-sized inputs to Einsum Op and improved its performance
    • Enable multi-worker NCCL all-reduce inside functions executing eagerly.
    • Added complex128 support to RFFT, RFFT2D, RFFT3D, IRFFT, IRFFT2D, and IRFFT3D.
    • Add pfor converter for SelfAdjointEigV2.
    • Add tf.math.ndtri and tf.math.erfinv.
    • Add tf.config.experimental.enable_mlir_bridge to allow using MLIR compiler bridge in eager model.
    • Added support for MatrixSolve on Cloud TPU / XLA.
    • Added tf.autodiff.ForwardAccumulator for forward-mode autodiff
    • Add LinearOperatorPermutation.
    • A few performance optimizations on tf.reduce_logsumexp.
    • Added multilabel handling to AUC metric
    • Optimization on zeros_like.
    • Dimension constructor now requires None or types with an __index__ method.
    • Add tf.random.uniform microbenchmark.
    • Use _protogen suffix for proto library targets instead of _cc_protogen suffix.
    • Moving the checkpoint reader from swig to pybind11.
    • tf.device & MirroredStrategy now supports passing in a tf.config.LogicalDevice
    • If you're building Tensorflow from source, consider using bazelisk to automatically download and use the correct Bazel version. Bazelisk reads the .bazelversion file at the root of the project directory.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

8bitmp3, Aaron Ma, AbdüLhamit Yilmaz, Abhai Kollara, aflc, Ag Ramesh, Albert Z. Guo, Alex Torres, amoitra, Andrii Prymostka, angeliand, Anshuman Tripathy, Anthony Barbier, Anton Kachatkou, Anubh-V, Anuja Jakhade, Artem Ryabov, autoih, Bairen Yi, Bas Aarts, Basit Ayantunde, Ben Barsdell, Bhavani Subramanian, Brett Koonce, candy.dc, Captain-Pool, caster, cathy, Chong Yan, Choong Yin Thong, Clayne Robison, Colle, Dan Ganea, David Norman, David Refaeli, dengziming, Diego Caballero, Divyanshu, djshen, Douman, Duncan Riach, EFanZh, Elena Zhelezina, Eric Schweitz, Evgenii Zheltonozhskii, Fei Hu, fo40225, Fred Reiss, Frederic Bastien, Fredrik Knutsson, fsx950223, fwcore, George Grzegorz Pawelczak, George Sterpu, Gian Marco Iodice, Giorgio Arena, giuros01, Gomathi Ramamurthy, Guozhong Zhuang, Haifeng Jin, Haoyu Wu, HarikrishnanBalagopal, HJYOO, Huang Chen-Yi, Ilham Firdausi Putra, Imran Salam, Jared Nielsen, Jason Zaman, Jasper Vicenti, Jeff Daily, Jeff Poznanovic, Jens Elofsson, Jerry Shih, jerryyin, Jesper Dramsch, jim.meyer, Jongwon Lee, Jun Wan, Junyuan Xie, Kaixi Hou, kamalkraj, Kan Chen, Karthik Muthuraman, Keiji Ariyama, Kevin Rose, Kevin Wang, Koan-Sin Tan, kstuedem, Kwabena W. Agyeman, Lakshay Tokas, latyas, Leslie-Fang-Intel, Li, Guizi, Luciano Resende, Lukas Folle, Lukas Geiger, Mahmoud Abuzaina, Manuel Freiberger, Mark Ryan, Martin Mlostek, Masaki Kozuki, Matthew Bentham, Matthew Denton, mbhuiyan, mdfaijul, Muhwan Kim, Nagy Mostafa, nammbash, Nathan Luehr, Nathan Wells, Niranjan Hasabnis, Oleksii Volkovskyi, Olivier Moindrot, olramde, Ouyang Jin, OverLordGoldDragon, Pallavi G, Paul Andrey, Paul Wais, pkanwar23, Pooya Davoodi, Prabindh Sundareson, Rajeshwar Reddy T, Ralovich, Kristof, Refraction-Ray, Richard Barnes, richardbrks, Robert Herbig, Romeo Kienzler, Ryan Mccormick, saishruthi, Saket Khandelwal, Sami Kama, Sana Damani, Satoshi Tanaka, Sergey Mironov, Sergii Khomenko, Shahid, Shawn Presser, ShengYang1, Siddhartha Bagaria, Simon Plovyt, skeydan, srinivasan.narayanamoorthy, Stephen Mugisha, sunway513, Takeshi Watanabe, Taylor Jakobson, TengLu, TheMindVirus, ThisIsIsaac, Tim Gates, Timothy Liu, Tomer Gafner, Trent Lo, Trevor Hickey, Trevor Morris, vcarpani, Wei Wang, Wen-Heng (Jack) Chung, wenshuai, Wenshuai-Xiaomi, wenxizhu, william, William D. Irons, Xinan Jiang, Yannic, Yasir Modak, Yasuhiro Matsumoto, Yong Tang, Yongfeng Gu, Youwei Song, Zaccharie Ramzi, Zhang, Zhenyu Guo, 王振华 (Zhenhua Wang), 韩董, 이중건 Isaac Lee

Release 1.15.0

This is the last 1.x release for TensorFlow. We do not expect to update the 1.x branch with features, although we will issue patch releases to fix vulnerabilities for at least one year.

Major Features and Improvements

  • As announced, tensorflow pip package will by default include GPU support (same as tensorflow-gpu now) for the platforms we currently have GPU support (Linux and Windows). It will work on machines with and without Nvidia GPUs. tensorflow-gpu will still be available, and CPU-only packages can be downloaded at tensorflow-cpu for users who are concerned about package size.
  • TensorFlow 1.15 contains a complete implementation of the 2.0 API in its compat.v2 module. It contains a copy of the 1.15 main module (without contrib) in the compat.v1 module. TensorFlow 1.15 is able to emulate 2.0 behavior using the enable_v2_behavior() function. This enables writing forward compatible code: by explicitly importing either tensorflow.compat.v1 or tensorflow.compat.v2, you can ensure that your code works without modifications against an installation of 1.15 or 2.0.
  • EagerTensor now supports numpy buffer interface for tensors.
  • Add toggles tf.enable_control_flow_v2() and tf.disable_control_flow_v2() for enabling/disabling v2 control flow.
  • Enable v2 control flow as part of tf.enable_v2_behavior() and TF2_BEHAVIOR=1.
  • AutoGraph translates Python control flow into TensorFlow expressions, allowing users to write regular Python inside tf.function-decorated functions. AutoGraph is also applied in functions used with tf.data, tf.distribute and tf.keras APIS.
  • Adds enable_tensor_equality(), which switches the behavior such that:
    • Tensors are no longer hashable.
    • Tensors can be compared with == and !=, yielding a Boolean Tensor with element-wise comparison results. This will be the default behavior in 2.0.

Breaking Changes

  • Tensorflow code now produces 2 different pip packages: tensorflow_core containing all the code (in the future it will contain only the private implementation) and tensorflow which is a virtual pip package doing forwarding to tensorflow_core (and in the future will contain only the public API of tensorflow). We don't expect this to be breaking, unless you were importing directly from the implementation.
  • TensorFlow 1.15 is built using devtoolset7 (GCC7) on Ubuntu 16. This may lead to ABI incompatibilities with extensions built against earlier versions of TensorFlow.
  • Deprecated the use of constraint= and .constraint with ResourceVariable.
  • tf.keras:
    • OMP_NUM_THREADS is no longer used by the default Keras config. To configure the number of threads, use tf.config.threading APIs.
    • tf.keras.model.save_model and model.save now defaults to saving a TensorFlow SavedModel.
    • keras.backend.resize_images (and consequently, keras.layers.Upsampling2D) behavior has changed, a bug in the resizing implementation was fixed.
    • Layers now default to float32, and automatically cast their inputs to the layer's dtype. If you had a model that used float64, it will probably silently use float32 in TensorFlow2, and a warning will be issued that starts with Layer "layer-name" is casting an input tensor from dtype float64 to the layer's dtype of float32. To fix, either set the default dtype to float64 with tf.keras.backend.set_floatx('float64'), or pass dtype='float64' to each of the Layer constructors. See tf.keras.layers.Layer for more information.
    • Some tf.assert_* methods now raise assertions at operation creation time (i.e. when this Python line executes) if the input tensors' values are known at that time, not during the session.run(). When this happens, a noop is returned and the input tensors are marked non-feedable. In other words, if they are used as keys in feed_dict argument to session.run(), an error will be raised. Also, because some assert ops don't make it into the graph, the graph structure changes. A different graph can result in different per-op random seeds when they are not given explicitly (most often).

Bug Fixes and Other Changes

  • tf.estimator:
    • tf.keras.estimator.model_to_estimator now supports exporting to tf.train.Checkpoint format, which allows the saved checkpoints to be compatible with model.load_weights.
    • Fix tests in canned estimators.
    • Expose Head as public API.
    • Fixes critical bugs that help with DenseFeatures usability in TF2
  • tf.data:
    • Promoting unbatch from experimental to core API.
    • Adding support for datasets as inputs to from_tensors and from_tensor_slices and batching and unbatching of nested datasets.
  • tf.keras:
    • tf.keras.estimator.model_to_estimator now supports exporting to tf.train.Checkpoint format, which allows the saved checkpoints to be compatible with model.load_weights.
    • Saving a Keras Model using tf.saved_model.save now saves the list of variables, trainable variables, regularization losses, and the call function.
    • Deprecated tf.keras.experimental.export_saved_model and tf.keras.experimental.function. Please use tf.keras.models.save_model(..., save_format='tf') and tf.keras.models.load_model instead.
    • Add an implementation=3 mode for tf.keras.layers.LocallyConnected2D and tf.keras.layers.LocallyConnected1D layers using tf.SparseTensor to store weights, allowing a dramatic speedup for large sparse models.
    • Enable the Keras compile API experimental_run_tf_function flag by default. This flag enables single training/eval/predict execution path. With this 1. All input types are converted to Dataset. 2. When distribution strategy is not specified this goes through the no-op distribution strategy path. 3. Execution is wrapped in tf.function unless run_eagerly=True is set in compile.
    • Raise error if batch_size argument is used when input is dataset/generator/keras sequence.
  • tf.lite
    • Add GATHER support to NN API delegate.
    • tflite object detection script has a debug mode.
    • Add delegate support for QUANTIZE.
    • Added evaluation script for COCO minival.
    • Add delegate support for QUANTIZED_16BIT_LSTM.
    • Converts hardswish subgraphs into atomic ops.
  • Add support for defaulting the value of cycle_length argument of tf.data.Dataset.interleave to the number of schedulable CPU cores.
  • parallel_for: Add converter for MatrixDiag.
  • Add narrow_range attribute to QuantizeAndDequantizeV2 and V3.
  • Added new op: tf.strings.unsorted_segment_join.
  • Add HW acceleration support for topK_v2.
  • Add new TypeSpec classes.
  • CloudBigtable version updated to v0.10.0.
  • Expose Head as public API.
  • Update docstring for gather to properly describe the non-empty batch_dims case.
  • Added tf.sparse.from_dense utility function.
  • Improved ragged tensor support in TensorFlowTestCase.
  • Makes the a-normal form transformation in Pyct configurable as to which nodes are converted to variables and which are not.
  • ResizeInputTensor now works for all delegates.
  • Add EXPAND_DIMS support to NN API delegate TEST: expand_dims_test
  • tf.cond emits a StatelessIf op if the branch functions are stateless and do not touch any resources.
  • tf.cond, tf.while and if and while in AutoGraph now accept a nonscalar predicate if has a single element. This does not affect non-V2 control flow.
  • tf.while_loop emits a StatelessWhile op if the cond and body functions are stateless and do not touch any resources.
  • Refactors code in Quant8 LSTM support to reduce TFLite binary size.
  • Add support of local soft device placement for eager op.
  • Add HW acceleration support for LogSoftMax.
  • Added a function nested_value_rowids for ragged tensors.
  • Add guard to avoid acceleration of L2 Normalization with input rank != 4
  • Add tf.math.cumulative_logsumexp operation.
  • Add tf.ragged.stack.
  • Fix memory allocation problem when calling AddNewInputConstantTensor.
  • Delegate application failure leaves interpreter in valid state.
  • Add check for correct memory alignment to MemoryAllocation::MemoryAllocation().
  • Extracts NNAPIDelegateKernel from nnapi_delegate.cc
  • Added support for FusedBatchNormV3 in converter.
  • A ragged to dense op for directly calculating tensors.
  • Fix accidental quadratic graph construction cost in graph-mode tf.gradients().

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

a6802739, Aaron Ma, Abdullah Selek, Abolfazl Shahbazi, Ag Ramesh, Albert Z. Guo, Albin Joy, Alex Itkes, Alex Sergeev, Alexander Pivovarov, Alexey Romanov, alhkad, Amit Srivastava, amoitra, Andrew Lihonosov, Andrii Prymostka, Anuj Rawat, Astropeak, Ayush Agrawal, Bairen Yi, Bas Aarts, Bastian Eichenberger, Ben Barsdell, Benjamin Peterson, bhack, Bharat Raghunathan, Bhavani Subramanian, Bryan Cutler, candy.dc, Cao Zongyan, Captain-Pool, Casper Da Costa-Luis, Chen Guoyin, Cheng Chang, chengchingwen, Chong Yan, Choong Yin Thong, Christopher Yeh, Clayne Robison, Coady, Patrick, Dan Ganea, David Norman, Denis Khalikov, Deven Desai, Diego Caballero, Duncan Dean, Duncan Riach, Dwight J Lyle, Eamon Ito-Fisher, eashtian3, EFanZh, ejot, Elroy Ashtian Jr, Eric Schweitz, Fangjun Kuang, Fei Hu, fo40225, formath, Fred Reiss, Frederic Bastien, Fredrik Knutsson, G. Hussain Chinoy, Gabriel, gehring, George Grzegorz Pawelczak, Gianluca Varisco, Gleb Popov, Greg Peatfield, Guillaume Klein, Gurpreet Singh, Gustavo Lima Chaves, haison, Haraldur TóMas HallgríMsson, HarikrishnanBalagopal, HåKon Sandsmark, I-Hong, Ilham Firdausi Putra, Imran Salam, Jason Zaman, Jason Zavaglia, jayhpark530, jefby, Jeff Daily, Jeffrey Poznanovic, Jekyll Lai, Jeroen BéDorf, Jerry Shih, jerryyin, jiakai, JiangXIAO, Joe Bowser, Joel Shapiro, Johan Gunnarsson, Jojimon Varghese, Joon, Josh Beal, Julian Niedermeier, Jun Wan, Junqin Zhang, Junyuan Xie, Justin Tunis, Kaixi Hou, Karl Lessard, Karthik Muthuraman, Kbhute-Ibm, khanhlvg, Koock Yoon, kstuedem, Kyuwon Kim, Lakshay Tokas, leike666666, leonard951, Leslie-Fang, Leslie-Fang-Intel, Li, Guizi, Lukas Folle, Lukas Geiger, Mahmoud Abuzaina, Manraj Singh Grover, Margaret Maynard-Reid, Mark Ryan, Matt Conley, Matthew Bentham, Matthew Denton, mbhuiyan, mdfaijul, Mei Jie, merturl, MichaelKonobeev, Michal W. Tarnowski, minds, mpppk, musikisomorphie, Nagy Mostafa, Nayana Thorat, Neil, Niels Ole Salscheider, Niklas SilfverströM, Niranjan Hasabnis, ocjosen, olramde, Pariksheet Pinjari, Patrick J. Lopresti, Patrik Gustavsson, per1234, PeterLee, Phan Van Nguyen Duc, Phillip Kravtsov, Pooya Davoodi, Pranav Marathe, Putra Manggala, Qingqing Cao, Rajeshwar Reddy T, Ramon ViñAs, Rasmus Diederichsen, Reuben Morais, richardbrks, robert, RonLek, Ryan Jiang, saishruthi, Saket Khandelwal, Saleem Abdulrasool, Sami Kama, Sana-Damani, Sergii Khomenko, Severen Redwood, Shubham Goyal, Sigrid Keydana, Siju Samuel, sleighsoft, smilu97, Son Tran, Srini511, srinivasan.narayanamoorthy, Sumesh Udayakumaran, Sungmann Cho, Tae-Hwan Jung, Taehoon Lee, Takeshi Watanabe, TengLu, terryky, TheMindVirus, ThisIsIsaac, Till Hoffmann, Timothy Liu, Tomer Gafner, Tongxuan Liu, Trent Lo, Trevor Morris, Uday Bondhugula, Vasileios Lioutas, vbvg2008, Vishnuvardhan Janapati, Vivek Suryamurthy, Wei Wang, Wen-Heng (Jack) Chung, wenxizhu, William D. Irons, winstonq, wyzhao, Xiaoming (Jason) Cui, Xinan Jiang, Xinping Wang, Yann-Yy, Yasir Modak, Yong Tang, Yongfeng Gu, Yuchen Ying, Yuxin Wu, zyeric, 王振华 (Zhenhua Wang)

Release 2.0.0

Major Features and Improvements

TensorFlow 2.0 focuses on simplicity and ease of use, featuring updates like:

  • Easy model building with Keras and eager execution.
  • Robust model deployment in production on any platform.
  • Powerful experimentation for research.
  • API simplification by reducing duplication and removing deprecated endpoints.

For details on best practices with 2.0, see the Effective 2.0 guide

For information on upgrading your existing TensorFlow 1.x models, please refer to our Upgrade and Migration guides. We have also released a collection of tutorials and getting started guides.

Highlights

  • TF 2.0 delivers Keras as the central high level API used to build and train models. Keras provides several model-building APIs such as Sequential, Functional, and Subclassing along with eager execution, for immediate iteration and intuitive debugging, and tf.data, for building scalable input pipelines. Checkout guide for additional details.
  • Distribution Strategy: TF 2.0 users will be able to use the tf.distribute.Strategy API to distribute training with minimal code changes, yielding great out-of-the-box performance. It supports distributed training with Keras model.fit, as well as with custom training loops. Multi-GPU support is available, along with experimental support for multi worker and Cloud TPUs. Check out the guide for more details.
  • Functions, not Sessions. The traditional declarative programming model of building a graph and executing it via a tf.Session is discouraged, and replaced with by writing regular Python functions. Using the tf.function decorator, such functions can be turned into graphs which can be executed remotely, serialized, and optimized for performance.
  • Unification of tf.train.Optimizers and tf.keras.Optimizers. Use tf.keras.Optimizers for TF2.0. compute_gradients is removed as public API, use GradientTape to compute gradients.
  • AutoGraph translates Python control flow into TensorFlow expressions, allowing users to write regular Python inside tf.function-decorated functions. AutoGraph is also applied in functions used with tf.data, tf.distribute and tf.keras APIs.
  • Unification of exchange formats to SavedModel. All TensorFlow ecosystem projects (TensorFlow Lite, TensorFlow JS, TensorFlow Serving, TensorFlow Hub) accept SavedModels. Model state should be saved to and restored from SavedModels.
  • API Changes: Many API symbols have been renamed or removed, and argument names have changed. Many of these changes are motivated by consistency and clarity. The 1.x API remains available in the compat.v1 module. A list of all symbol changes can be found here.
    • API clean-up, included removing tf.app, tf.flags, and tf.logging in favor of absl-py.
  • No more global variables with helper methods like tf.global_variables_initializer and tf.get_global_step.
  • Add toggles tf.enable_control_flow_v2() and tf.disable_control_flow_v2() for enabling/disabling v2 control flow.
  • Enable v2 control flow as part of tf.enable_v2_behavior() and TF2_BEHAVIOR=1.
  • Fixes autocomplete for most TensorFlow API references by switching to use relative imports in API __init__.py files.
  • Auto Mixed-Precision graph optimizer simplifies converting models to float16 for acceleration on Volta and Turing Tensor Cores. This feature can be enabled by wrapping an optimizer class with tf.train.experimental.enable_mixed_precision_graph_rewrite().
  • Add environment variable TF_CUDNN_DETERMINISTIC. Setting to "true" or "1" forces the selection of deterministic cuDNN convolution and max-pooling algorithms. When this is enabled, the algorithm selection procedure itself is also deterministic.

Breaking Changes

  • Many backwards incompatible API changes have been made to clean up the APIs and make them more consistent.

  • Toolchains:

    • TensorFlow 2.0.0 is built using devtoolset7 (GCC7) on Ubuntu 16. This may lead to ABI incompatibilities with extensions built against earlier versions of TensorFlow.
    • Tensorflow code now produces 2 different pip packages: tensorflow_core containing all the code (in the future it will contain only the private implementation) and tensorflow which is a virtual pip package doing forwarding to tensorflow_core (and in the future will contain only the public API of tensorflow). We don't expect this to be breaking, unless you were importing directly from the implementation. Removed the freeze_graph command line tool; SavedModel should be used in place of frozen graphs.
  • tf.contrib:

    • tf.contrib has been deprecated, and functionality has been either migrated to the core TensorFlow API, to an ecosystem project such as tensorflow/addons or tensorflow/io, or removed entirely.
    • Remove tf.contrib.timeseries dependency on TF distributions.
    • Replace contrib references with tf.estimator.experimental.* for apis in early_stopping.py.
  • tf.estimator:

    • Premade estimators in the tf.estimator.DNN/Linear/DNNLinearCombined family have been updated to use tf.keras.optimizers instead of the tf.compat.v1.train.Optimizers. If you do not pass in an optimizer= arg or if you use a string, the premade estimator will use the Keras optimizer. This is checkpoint breaking, as the optimizers have separate variables. A checkpoint converter tool for converting optimizers is included with the release, but if you want to avoid any change, switch to the v1 version of the estimator: tf.compat.v1.estimator.DNN/Linear/DNNLinearCombined*.
    • Default aggregation for canned Estimators is now SUM_OVER_BATCH_SIZE. To maintain previous default behavior, please pass SUM as the loss aggregation method.
    • Canned Estimators don’t support input_layer_partitioner arg in the API. If you have this arg, you will have to switch to tf.compat.v1 canned Estimators.
    • Estimator.export_savedmodel has been renamed to export_saved_model.
    • When saving to SavedModel, Estimators will strip default op attributes. This is almost always the correct behavior, as it is more forwards compatible, but if you require that default attributes to be saved with the model, please use tf.compat.v1.Estimator.
    • Feature Columns have been upgraded to be more Eager-friendly and to work with Keras. As a result, tf.feature_column.input_layer has been deprecated in favor of tf.keras.layers.DenseFeatures. v1 feature columns have direct analogues in v2 except for shared_embedding_columns, which are not cross-compatible with v1 and v2. Use tf.feature_column.shared_embeddings instead.
  • tf.keras:

    • OMP_NUM_THREADS is no longer used by the default Keras config. To configure the number of threads, use tf.config.threading APIs.
    • tf.keras.model.save_model and model.save now defaults to saving a TensorFlow SavedModel. HDF5 files are still supported.
    • Deprecated tf.keras.experimental.export_saved_model and tf.keras.experimental.function. Please use tf.keras.models.save_model(..., save_format='tf') and tf.keras.models.load_model instead.
    • Layers now default to float32, and automatically cast their inputs to the layer's dtype. If you had a model that used float64, it will probably silently use float32 in TensorFlow 2, and a warning will be issued that starts with Layer <layer-name> is casting an input tensor from dtype float64 to the layer's dtype of float32. To fix, either set the default dtype to float64 with tf.keras.backend.set_floatx('float64'), or pass dtype='float64' to each of the Layer constructors. See tf.keras.layers.Layer for more information.
  • tf.lite:

    • Removed lite.OpHint, lite.experimental, and lite.constant from 2.0 API.
  • Tensors are no longer hashable, but instead compare element-wise with == and !=. Use tf.compat.v1.disable_tensor_equality() to return to the previous behavior.

  • Performing equality operations on Tensors or Variables with incompatible shapes an exception is no longer thrown. Instead __eq__ returns False and __ne__ returns True.

  • Removed tf.string_split from v2 API.

  • Deprecated the use of constraint= and .constraint with ResourceVariable.

  • Add UnifiedGRU as the new GRU implementation for tf2.0. Change the default recurrent activation function for GRU from hard_sigmoid to sigmoid, and reset_after to True in 2.0. Historically recurrent activation is hard_sigmoid since it is fast than 'sigmoid'. With new unified backend between CPU and GPU mode, since the CuDNN kernel is using sigmoid, we change the default for CPU mode to sigmoid as well. With that, the default GRU will be compatible with both CPU and GPU kernel. This will enable user with GPU to use CuDNN kernel by default and get a 10x performance boost in training. Note that this is checkpoint breaking change. If user want to use their 1.x pre-trained checkpoint, please construct the layer with GRU(recurrent_activation='hard_sigmoid', reset_after=False) to fallback to 1.x behavior.

  • CUDNN_INSTALL_PATH, TENSORRT_INSTALL_PATH, NCCL_INSTALL_PATH, NCCL_HDR_PATH are deprecated. Use TF_CUDA_PATHS instead which supports a comma-separated list of base paths that are searched to find CUDA libraries and headers.

Refer to our public project status tracker and issues tagged with 2.0 on GitHub for insight into recent issues and development progress.

If you experience any snags when using TF 2.0, please let us know at the TF 2.0 Testing User Group. We have a support mailing list as well as weekly testing meetings, and would love to hear your migration feedback and questions.

Bug Fixes and Other Changes

  • tf.contrib:

    • Expose tf.contrib.proto.* ops in tf.io (they will exist in TF2)
  • tf.data:

    • Add support for TensorArrays to tf.data Dataset.
    • Integrate Ragged Tensors with tf.data.
    • All core and experimental tf.data transformations that input user-defined functions can span multiple devices now.
    • Extending the TF 2.0 support for shuffle(..., reshuffle_each_iteration=True) and cache() to work across different Python iterators for the same dataset.
    • Removing the experimental_numa_aware option from tf.data.Options.
    • Add num_parallel_reads and passing in a Dataset containing filenames into TextLineDataset and FixedLengthRecordDataset.
    • Add support for defaulting the value of cycle_length argument of tf.data.Dataset.interleave to the number of schedulable CPU cores.
    • Promoting tf.data.experimental.enumerate_dataset to core as tf.data.Dataset.enumerate.
    • Promoting tf.data.experimental.unbatch to core as tf.data.Dataset.unbatch.
    • Adds option for introducing slack in the pipeline to reduce CPU contention, via tf.data.Options().experimental_slack = True
    • Added experimental support for parallel batching to batch() and padded_batch(). This functionality can be enabled through tf.data.Options().
    • Support cancellation of long-running reduce.
    • Now we use dataset node name as prefix instead of the op name, to identify the component correctly in metrics, for pipelines with repeated components.
    • Improve the performance of datasets using from_tensors().
    • Promoting unbatch from experimental to core API.
    • Adding support for datasets as inputs to from_tensors and from_tensor_slices and batching and unbatching of nested datasets.
  • tf.distribute:

    • Enable tf.distribute.experimental.MultiWorkerMirroredStrategy working in eager mode.
    • Callbacks are supported in MultiWorkerMirroredStrategy.
    • Disable run_eagerly and distribution strategy if there are symbolic tensors added to the model using add_metric or add_loss.
    • Loss and gradients should now more reliably be correctly scaled w.r.t. the global batch size when using a tf.distribute.Strategy.
    • Set default loss reduction as AUTO for improving reliability of loss scaling with distribution strategy and custom training loops. AUTO indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used in distribution strategy scope, outside of built-in training loops such as tf.keras compile and fit, we expect reduction value to be 'None' or 'SUM'. Using other values will raise an error.
    • Support for multi-host ncclAllReduce in Distribution Strategy.
  • tf.estimator:

    • Replace tf.contrib.estimator.add_metrics with tf.estimator.add_metrics
    • Use tf.compat.v1.estimator.inputs instead of tf.estimator.inputs
    • Replace contrib references with tf.estimator.experimental.* for apis in early_s in Estimator
    • Canned Estimators will now use keras optimizers by default. An error will be raised if tf.train.Optimizers are used, and you will have to switch to tf.keras.optimizers or tf.compat.v1 canned Estimators.
    • A checkpoint converter for canned Estimators has been provided to transition canned Estimators that are warm started from tf.train.Optimizers to tf.keras.optimizers.
    • Losses are scaled in canned estimator v2 and not in the optimizers anymore. If you are using Estimator + distribution strategy + optimikzer v1 then the behavior does not change. This implies that if you are using custom estimator with optimizer v2, you have to scale losses. We have new utilities to help scale losses tf.nn.compute_average_loss, tf.nn.scale_regularization_loss.
  • tf.keras:

    • Premade models (including Linear and WideDeep) have been introduced for the purpose of replacing Premade estimators.
    • Model saving changes
    • model.save and tf.saved_model.save may now save to the TensorFlow SavedModel format. The model can be restored using tf.keras.models.load_model. HDF5 files are still supported, and may be used by specifying save_format="h5" when saving.
    • Raw TensorFlow functions can now be used in conjunction with the Keras Functional API during model creation. This obviates the need for users to create Lambda layers in most cases when using the Functional API. Like Lambda layers, TensorFlow functions that result in Variable creation or assign ops are not supported.
    • Add support for passing list of lists to the metrics argument in Keras compile.
    • Add tf.keras.layers.AbstractRNNCell as the preferred implementation for RNN cells in TF v2. User can use it to implement RNN cells with custom behavior.
    • Keras training and validation curves are shown on the same plot when using the TensorBoard callback.
    • Switched Keras fit/evaluate/predict execution to use only a single unified path by default unless eager execution has been explicitly disabled, regardless of input type. This unified path places an eager-friendly training step inside of a tf.function. With this
    • All input types are converted to Dataset.
    • The path assumes there is always a distribution strategy. when distribution strategy is not specified the path uses a no-op distribution strategy.
    • The training step is wrapped in tf.function unless run_eagerly=True is set in compile. The single path execution code does not yet support all use cases. We fallback to the existing v1 execution paths if your model contains the following:
      1. sample_weight_mode in compile
      2. weighted_metrics in compile
      3. v1 optimizer
      4. target tensors in compile If you are experiencing any issues because of this change, please inform us (file an issue) about your use case and you can unblock yourself by setting experimental_run_tf_function=False in compile meanwhile. We have seen couple of use cases where the model usage pattern is not as expected and would not work with this change.
    • output tensors of one layer is used in the constructor of another.
    • symbolic tensors outside the scope of the model are used in custom loss functions. The flag can be disabled for these cases and ideally the usage pattern will need to be fixed.
    • Mark Keras set_session as compat.v1 only.
    • tf.keras.estimator.model_to_estimator now supports exporting to tf.train.Checkpoint format, which allows the saved checkpoints to be compatible with model.load_weights.
    • keras.backend.resize_images (and consequently, keras.layers.Upsampling2D) behavior has changed, a bug in the resizing implementation was fixed.
    • Add an implementation=3 mode for tf.keras.layers.LocallyConnected2D and tf.keras.layers.LocallyConnected1D layers using tf.SparseTensor to store weights, allowing a dramatic speedup for large sparse models.
    • Raise error if batch_size argument is used when input is dataset/generator/keras sequence.
    • Update TF 2.0 keras.backend.name_scope to use TF 2.0 name_scope.
    • Add v2 module aliases for losses, metrics, initializers and optimizers: tf.losses = tf.keras.losses & tf.metrics = tf.keras.metrics & tf.initializers = tf.keras.initializers & tf.optimizers = tf.keras.optimizers.
    • Updates binary cross entropy logic in Keras when input is probabilities. Instead of converting probabilities to logits, we are using the cross entropy formula for probabilities.
    • Added public APIs for cumsum and cumprod keras backend functions.
    • Add support for temporal sample weight mode in subclassed models.
    • Raise ValueError if an integer is passed to the training APIs.
    • Added fault-tolerance support for training Keras model via model.fit() with MultiWorkerMirroredStrategy, tutorial available.
    • Custom Callback tutorial is now available.
    • To train with tf.distribute, Keras API is recommended over estimator.
    • steps_per_epoch and steps arguments are supported with numpy arrays.
    • New error message when unexpected keys are used in sample_weight/class_weight dictionaries
    • Losses are scaled in Keras compile/fit and not in the optimizers anymore. If you are using custom training loop, we have new utilities to help scale losses tf.nn.compute_average_loss, tf.nn.scale_regularization_loss.
    • Layer apply and add_variable APIs are deprecated.
    • Added support for channels first data format in cross entropy losses with logits and support for tensors with unknown ranks.
    • Error messages will be raised if add_update, add_metric, add_loss, activity regularizers are used inside of a control flow branch.
    • New loss reduction types:
    • AUTO: Indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used with tf.distribute.Strategy, outside of built-in training loops such as tf.keras compile and fit, we expect reduction value to be SUM or NONE. Using AUTO in that case will raise an error.
    • NONE: Weighted losses with one dimension reduced (axis=-1, or axis specified by loss function). When this reduction type used with built-in Keras training loops like fit/evaluate, the unreduced vector loss is passed to the optimizer but the reported loss will be a scalar value.
    • SUM: Scalar sum of weighted losses. 4. SUM_OVER_BATCH_SIZE: Scalar SUM divided by number of elements in losses. This reduction type is not supported when used with tf.distribute.Strategy outside of built-in training loops like tf.keras compile/fit.
    • Wraps losses passed to the compile API (strings and v1 losses) which are not instances of v2 Loss class in LossWrapper class. => All losses will now use SUM_OVER_BATCH_SIZE reduction as default.
    • model.add_loss(symbolic_tensor) should work in ambient eager.
    • Update metric name to always reflect what the user has given in compile. Affects following cases
    • When name is given as 'accuracy'/'crossentropy'
    • When an aliased function name is used eg. 'mse'
    • Removing the weighted prefix from weighted metric names.
    • Allow non-Tensors through v2 losses.
    • Add v2 sparse categorical crossentropy metric.
    • Add v2 APIs for AUCCurve and AUCSummationMethod enums.
    • add_update can now be passed a zero-arg callable in order to support turning off the update when setting trainable=False on a Layer of a Model compiled with run_eagerly=True.
    • Standardize the LayerNormalization API by replacing the args norm_axis and params_axis with axis.
    • Fixed critical bugs that help with DenseFeatures usability in TF2
  • tf.lite:

    • Added evaluation script for COCO minival
    • Add delegate support for QUANTIZE.
    • Add GATHER support to NN API delegate.
    • Added support for TFLiteConverter Python API in 2.0. Contains functions from_saved_model, from_keras_file, and from_concrete_functions.
    • Add EXPAND_DIMS support to NN API delegate TEST.
    • Add narrow_range attribute to QuantizeAndDequantizeV2 and V3.
    • Added support for tflite_convert command line tool in 2.0.
    • Post-training quantization tool supports quantizing weights shared by multiple operations. The models made with versions of this tool will use INT8 types for weights and will only be executable interpreters from this version onwards.
    • Post-training quantization tool supports fp16 weights and GPU delegate acceleration for fp16.
    • Add delegate support for QUANTIZED_16BIT_LSTM.
    • Extracts NNAPIDelegateKernel from nnapi_delegate.cc
  • TensorRT

    • Add TensorFlow 2.0-compatible TrtGraphConverterV2 API for TensorRT conversion. TensorRT initialization arguments are now passed wrapped in a named-tuple, TrtConversionParams, rather than as separate arguments as in TrtGraphConverter.
    • Changed API to optimize TensorRT engines during graph optimization. This is now done by calling converter.build() where previously is_dynamic_op=False would be set.
    • converter.convert() no longer returns a tf.function. Now the function must be accessed from the saved model.
    • The converter.calibrate() method has been removed. To trigger calibration, a calibration_input_fn should be provided to converter.convert().
  • Other:

    • Fix accidental quadratic graph construction cost in graph-mode tf.gradients().
    • ResourceVariable's gather op supports batch dimensions.
    • ResourceVariable support for gather_nd.
    • ResourceVariable and Variable no longer accepts constraint in the constructor, nor expose it as a @property.
    • Added gradient for SparseToDense op.
    • Expose a flag that allows the number of threads to vary across Python benchmarks.
    • image.resize in 2.0 now supports gradients for the new resize kernels.
    • image.resize now considers proper pixel centers and has new kernels (incl. anti-aliasing).
    • Renamed tf.image functions to remove duplicate "image" where it is redundant.
    • Variadic reduce is supported on CPU Variadic reduce is supported on CPU
    • Remove unused StringViewVariantWrapper.
    • Delete unused Fingerprint64Map op registration
    • Add broadcasting support to tf.matmul.
    • Add C++ Gradient for BatchMatMulV2.
    • Add tf.math.cumulative_logsumexp operation.
    • Add ellipsis (...) support for tf.einsum().
    • Add expand_composites argument to all nest.* methods.
    • Added strings.byte_split.
    • Add a new "result_type" parameter to tf.strings.split.
    • Add name argument to tf.string_split and tf.strings_split.
    • Extend tf.strings.split to support inputs with any rank.
    • Added tf.random.binomial.
    • Added key and skip methods to random.experimental.Generator.
    • Extend tf.function with basic support for CompositeTensors arguments (such as SparseTensor and RaggedTensor).
    • parallel_for.pfor: add converters for Softmax, LogSoftmax, IsNaN, All, Any, and MatrixSetDiag.
    • parallel_for: add converters for LowerTriangularSolve and Cholesky.
    • parallel_for: add converters for LogMatrixDeterminant and MatrixBandPart.
    • parallel_for: Add converter for MatrixDiag.
    • parallel_for: Add converters for OneHot, LowerBound, UpperBound.
    • parallel_for: add converter for BroadcastTo.
    • Add pfor converter for Squeeze.
    • Add RaggedTensor.placeholder().
    • Add ragged tensor support to tf.squeeze.
    • Update RaggedTensors to support int32 row_splits.
    • Allow LinearOperator.solve to take a LinearOperator.
    • Allow all dtypes for LinearOperatorCirculant.
    • Introduce MaxParallelism method
    • Add LinearOperatorHouseholder.
    • Adds Philox support to new stateful RNG's XLA path.
    • Added TensorSpec support for CompositeTensors.
    • Added tf.linalg.tridiagonal_solve op.
    • Added partial_pivoting input parameter to tf.linalg.tridiagonal_solve.
    • Added gradient to tf.linalg.tridiagonal_solve.
    • Added tf.linalg.tridiagonal_mul op.
    • Added GPU implementation of tf.linalg.tridiagonal_matmul.
    • Added LinearOperatorToeplitz.
    • Upgraded LIBXSMM to version 1.11.
    • Uniform processing of quantized embeddings by Gather and EmbeddingLookup Ops.
    • Correct a misstatement in the documentation of the sparse softmax cross entropy logit parameter.
    • Add tf.ragged.boolean_mask.
    • tf.switch_case added, which selects a branch_fn based on a branch_index.
    • The C++ kernel of gather op supports batch dimensions.
    • Fixed default value and documentation for trainable arg of tf.Variable.
    • EagerTensor now supports numpy buffer interface for tensors.
    • This change bumps the version number of the FullyConnected Op to 5.
    • Added new op: tf.strings.unsorted_segment_join.
    • Added HW acceleration support for topK_v2.
    • CloudBigtable version updated to v0.10.0 BEGIN_PUBLIC CloudBigtable version updated to v0.10.0.
    • Expose Head as public API.
    • Added tf.sparse.from_dense utility function.
    • Improved ragged tensor support in TensorFlowTestCase.
    • Added a function nested_value_rowids for ragged tensors.
    • Added tf.ragged.stack.
    • Makes the a-normal form transformation in Pyct configurable as to which nodes are converted to variables and which are not.
    • ResizeInputTensor now works for all delegates.
    • tf.cond emits a StatelessIf op if the branch functions are stateless and do not touch any resources.
    • Add support of local soft device placement for eager op.
    • Pass partial_pivoting to the _TridiagonalSolveGrad.
    • Add HW acceleration support for LogSoftMax.
    • Add guard to avoid acceleration of L2 Normalization with input rank != 4
    • Fix memory allocation problem when calling AddNewInputConstantTensor.
    • Delegate application failure leaves interpreter in valid state
    • tf.while_loop emits a StatelessWhile op if the cond and body functions are stateless and do not touch any resources.
    • tf.cond, tf.while and if and while in AutoGraph now accept a nonscalar predicate if has a single element. This does not affect non-V2 control flow.
    • Fix potential security vulnerability where decoding variant tensors from proto could result in heap out of bounds memory access.
    • Only create a GCS directory object if the object does not already exist.
    • Introduce dynamic constructor argument in Layer and Model, which should be set to True when using imperative control flow in the call method.
    • Begin adding Go wrapper for C Eager API.
    • XLA HLO graphs can be inspected with interactive_graphviz tool now.
    • Add dataset ops to the graph (or create kernels in Eager execution) during the python Dataset object creation instead doing it during Iterator creation time.
    • Add batch_dims argument to tf.gather.
    • The behavior of tf.gather is now correct when axis=None and batch_dims<0.
    • Update docstring for gather to properly describe the non-empty batch_dims case.
    • Removing of dtype in the constructor of initializers and partition_info in call.
    • Add tf.math.nextafter op.
    • Turn on MKL-DNN contraction kernels by default. MKL-DNN dynamically dispatches the best kernel implementation based on CPU vector architecture. To disable them, build with --define=tensorflow_mkldnn_contraction_kernel=0.
    • tf.linspace(start, stop, num) now always uses "stop" as last value (for num > 1)
    • Added top-k to precision and recall to keras metrics.
    • Add a ragged size op and register it to the op dispatcher
    • Transitive dependencies on :pooling_ops were removed. Some users may need to add explicit dependencies on :pooling_ops if they reference the operators from that library.
    • Add CompositeTensor base class.
    • Malformed gif images could result in an access out of bounds in the color palette of the frame. This has been fixed now
    • Add templates and interfaces for creating lookup tables
    • Tensor::UnsafeCopyFromInternal deprecated in favor Tensor::BitcastFrom.
    • In map_vectorization optimization, reduce the degree of parallelism in the vectorized map node.
    • Add variant wrapper for absl::string_view.
    • Add OpKernels for some stateless maps.
    • DType is no longer convertible to an int. Use dtype.as_datatype_enum instead of int(dtype) to get the same result.
    • Support both binary and -1/1 label input in v2 hinge and squared hinge losses.
    • Added LinearOperator.adjoint and LinearOperator.H (alias).
    • Expose CriticalSection in core as tf.CriticalSection.
    • Enhanced graphviz output.
    • Add opkernel templates for common table operations.
    • Fix callbacks do not log values in eager mode when a deferred build model is used.
    • SignatureDef util functions have been deprecated.
    • Update Fingerprint64Map to use aliases
    • Add legacy string flat hash map op kernels.
    • Add support for add_metric in the graph function mode.
    • Updating cosine similarity loss - removed the negate sign from cosine similarity.
    • Changed default for gradient accumulation for TPU embeddings to true.
    • Adds summary trace API for collecting graph and profile information.
    • The precision_mode argument to TrtGraphConverter is now case insensitive.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

1e100, a6802739, 4d55397500, a6802739, Abdullah Selek, abenmao, Abolfazl Shahbazi, Adam Richter, Adam Weiss, Ag Ramesh, Alan Du, Albin Joy, Alex, Alex Itkes, Alex Sergeev, Alexander Pivovarov, Alexey Romanov, alhkad, Aman Patel, Amit, Amit Kumar Jaiswal, Amit Srivastava, amoitra, Andreas Eberle, Andrew Lihonosov, Andy Craze, Anshuman Tripathy, Anthony Hsu, Anthony Platanios, Anuj Rawat, arp95, Arpit Shah, Armen Poghosov, armenpoghosov, Astropeak, Ashwin Ramaswami, Arpit Shah, Augustina Ragwitz, Aurelien Geron, AuréLien Geron, avasid, aweers, awesomealex1, Ayush Agrawal, Bas Aarts, Bastian Eichenberger, Bairen Yi, Bayberry Z, Ben Barsdell, Benjamin Peterson, bhack, Bharat Raghunathan, Bhavani Subramanian, Bin Fan, blairhan, BléNesi Attila, Bodin-E, Brandon Carter, Bryan Cutler, candy.dc, Cao Zongyan, Casper Da Costa-Luis, Chao Liu, Chen Guoyin, chenchc, chengchingwen, chie8842, Christian Hansen, Christoph Boeddeker, Christopher Yeh, Clayne Robison, Coady, Patrick, crafet, csukuangfj, ctiijima, Dan Jarvis, Dan Lazewatsky, Daniel Ingram, Daniel Rasmussen, Daniel Salvadori, Dave Airlie, David Norman, Dayananda V, delock, Denis Khalikov, Deven Desai, Dheeraj Rajaram Reddy, Diego Caballero, dmitrievanthony, Donovan Ong, Drew Szurko, Duncan Dean, Duncan Riach, Dustin Neighly, Dwight J Lyle, Eamon Ito-Fisher, eashtian3, Edward Forgacs, EFanZh, ejot, Elroy Ashtian Jr, Eric Schweitz, Evgeniy Polyakov, Fangjun Kuang, Federico Martinez, Fei Hu, Felix Lemke, Filip Matzner, FlashTek, fo40225, formath, FrançOis Chollet, frreiss, Fred Reiss, Frederic Bastien, Fredrik Knutsson, G. Hussain Chinoy, Gabriel, Gautam, gehring, Geoffrey Irving, George Grzegorz Pawelczak, Grzegorz Pawelczak, George Sterpu, Gianluca Varisco, Gleb Popov, Greg Peatfield, Guillaume Klein, Gurpreet Singh, Gustavo Lima Chaves, Gyoung-Yoon Ryoo, haison, Hanton Yang, HanGuo97, Haraldur TóMas HallgríMsson, Hari Shankar, hehongliang, Heungsub Lee, Hoeseong Kim, Huan Li (李卓桓), HåKon Sandsmark, I-Hong, I-Hong Jhuo, Ilham Firdausi Putra, Ilango R, Imran Salam, Innovimax, Jacky Ko, Irene Dea, Ivan Habernal, Jakub Lipinski, Jacky, Jason Zaman, Jason Zavaglia, jayhpark530, jcf94, jefby, Jeff Daily, Jeff Poznanovic, Jeffrey Poznanovic, Jekyll Lai, jer, Jeroen BéDorf, jerryyin, jhalakp, jiakai, Jia Qingtong, Jiankang, JiangXIAO, Joe Bowser, Joe Q, Joe Quadrino, Joel Shapiro, Johan Gunnarsson, Jojimon Varghese, Jonas Rauber, Jonathan Kyl, Jonathan, Joon, Joppe Geluykens, Joseph Friedman, Josh Beal, jtressle, Julian Niedermeier, Junqin Zhang, Justin Dujardin, Justin Tunis, jwu, K. Hodges, kaixih, Kaixi Hou, kjopek, Karl Lessard, Karl Weinmeister, Karthik Muthuraman, Kashif Rasul, Kay Zhu, Kbhute-Ibm, KDR, Keno Fischer, Kevin Mader, khanhlvg, Kilaru Yasaswi Sri Chandra Gandhi, Koan-Sin Tan, Koock Yoon, kouml, ktaebum, Kyuwon Kim, Lakshay Tokas, Laurent Le Brun, leike666666, leonard951, Leslie-Fang, Letian Kang, Li, Guizi, Loo Rong Jie, Lucas Hendren, Lukas Folle, Lukas Geiger, Luke Han, luxupu, lvli, Ma, Guokai, Mahmoud Abuzaina, Maksym Kysylov, Mandar Deshpande, manhyuk, Manraj Singh Grover, Marco Gaido, Marek Drozdowski, Margaret Maynard-Reid, Mark Ryan, mars20, Mateusz Chudyk, Matt Conley, mbhuiyan, mdfaijul, Mei Jie, Melissa Grueter, merturl, MichaelKonobeev, Michael KäUfl, Michal W. Tarnowski, MickaëL Schoentgen, Miguel Morin, Mihail Salnikov, Mikalai Drabovich, Mike Arpaia, Mike Holcomb, minds, monklof, Moses Marin, mpppk, Mr. Metal, Mshr-H, musikisomorphie, nammbash, Natalia Gimelshein, Nathan Luehr, Nayana-Ibm, Nayana Thorat, neargye, Neeraj Pradhan, Nehal J Wani, Neil, Nick, Nick Lewycky, Niels Ole Salscheider, Niklas SilfverströM, Niranjan Hasabnis, Nuka-137, Nutti, ocjosen, olicht, omeir1, P Sudeepam, Paige Bailey, Palmer Lao, Pan Daoxin, Pariksheet Pinjari, Pasquale Minervini, Patrick J. Lopresti, Patrik Gustavsson, Pavel Akhtyamov, Pavel Samolysov, PENGWA, per1234, PeterLee, Phan Van Nguyen Duc, Philipp Jund, Phillip Kravtsov, Pooya Davoodi, Pranav Marathe, Putra Manggala, Qingqing Cao, R S Nikhil Krishna, Rajeshwar Reddy T, Ramon ViñAs, Rasmus Diederichsen, Reuben Morais, robert, Rohit Gupta, Roland Zimmermann, Roman Soldatow, RonLek, Ruizhe, Ryan Jiang, saishruthi, Saleem Abdulrasool, Samantha Andow, Sami Kama, Sana-Damani, Saurabh Deoras, sdamani, Sean Morgan, seanshpark, Sebastien Iooss, Serv-Inc, Severen Redwood, Shahzad Lone, Shashank Gupta, shashvat, Shashvat Chand Shahi, Shubham Goyal, Shashi, Sigrid Keydana, Siju, Siju Samuel, sleighsoft, smilu97, Snease-Abq, Son Tran, Spencer Schaber, sremedios, Srini511, srinivasan.narayanamoorthy, Steve Lang, Steve Nesae, Subin, Sumesh Udayakumaran, Sungmann Cho, sunway513, Supriya Rao, sxwang, Tae-Hwan Jung, Taehoon Lee, Takeo Sawada, Taylor Jakobson, Taylor Thornton, Ted Chang, TengLu, terryky, ThisIsIsaac, ThisIsPIRI, Thomas Deegan, Thomas Hagebols, tianyapiaozi, Till Hoffmann, Tim Zaman, tomguluson92, Tongxuan Liu, Trent Lo, Trevor Morris, TungJerry, Tyorden, Uday Bondhugula, v1incent, Vagif, Vasileios Lioutas, vbvg2008, vcarpani, Vijay Ravichandran, Vikram Tiwari,Viktor Gal, Vishwak Srinivasan, Vincent, Vishnuvardhan Janapati, Vitor-Alves, Vivek Suryamurthy, wangsiyu, wateryzephyr, WeberXie, Wei Wang, WeijieSun, Wen-Heng (Jack) Chung, wenxizhu, Will Battel, William D. Irons, winstonq, wyzhao, Xiaoming (Jason) Cui, Xiaoquan Kong, Xin, Xinping Wang, Yan Facai (颜发才), Yann-Yy, Yasir Modak, Yasuhiro Matsumoto, ymodak, Yong Tang, Yongfeng Gu, Younes Khoudli, Yuan Lin, Yuan (Terry) Tang, Yuchen Ying, Yves-Noel Weweler, zhangyujing, zjjott, zyeric, 王振华 (Zhenhua Wang), 黄鑫

Release 1.14.0

Major Features and Improvements

  • This is the first 1.x release containing the compat.v2 module. This module is required to allow libraries to publish code which works in both 1.x and 2.x. After this release, no backwards incompatible changes are allowed in the 2.0 Python API.
  • Turn on MKL-DNN contraction kernels by default. MKL-DNN dynamically dispatches the best kernel implementation based on CPU vector architecture. To disable them, build with --define=tensorflow_mkldnn_contraction_kernel=0.

Behavioral changes

  • Set default loss reduction as AUTO for improving reliability of loss scaling with distribution strategy and custom training loops. AUTO indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used in distribution strategy scope, outside of built-in training loops such as tf.keras compile and fit, we expect reduction value to be 'None' or 'SUM'. Using other values will raise an error.
  • Wraps losses passed to the compile API (strings and v1 losses) which are not instances of v2 Loss class in LossWrapper class. => All losses will now use SUM_OVER_BATCH_SIZE reduction as default.
  • Disable run_eagerly and distribution strategy if there are symbolic tensors added to the model using add_metric or add_loss.
  • tf.linspace(start, stop, num) now always uses "stop" as last value (for num > 1)
  • ResourceVariable and Variable no longer accepts constraint in the constructor, nor expose it as a @property.
  • The behavior of tf.gather is now correct when axis=None and batch_dims<0.
  • Only create a GCS directory object if the object does not already exist.
  • In map_vectorization optimization, reduce the degree of parallelism in the vectorized map node.
  • Bug fix: loss and gradients should now more reliably be correctly scaled w.r.t. the global batch size when using a tf.distribute.Strategy.
  • Updating cosine similarity loss - removed the negate sign from cosine similarity.
  • DType is no longer convertible to an int. Use dtype.as_datatype_enum instead of int(dtype) to get the same result.
  • Changed default for gradient accumulation for TPU embeddings to true.
  • Callbacks now log values in eager mode when a deferred build model is used.
  • Transitive dependencies on :pooling_ops were removed. Some users may need to add explicit dependencies on :pooling_ops if they reference the operators from that library.
  • tf.keras.optimizers default learning rate changes:
    • Adadelta: 1.000 to 0.001
    • Adagrad: 0.01 to 0.001
    • Adamax: 0.002 to 0.001
    • NAdam: 0.002 to 0.001

Bug Fixes and Other Changes

  • Documentation
  • Deprecations and Symbol renames.
    • Remove unused StringViewVariantWrapper
    • Delete unused Fingerprint64Map op registration
    • SignatureDef util functions have been deprecated.
    • Renamed tf.image functions to remove duplicate "image" where it is redundant.
    • tf.keras.experimental.export renamed to tf.keras.experimental.export_saved_model
    • Standardize the LayerNormalization API by replacing the args norm_axis and params_axis with axis.
    • Tensor::UnsafeCopyFromInternal deprecated in favor Tensor::BitcastFrom
  • Keras & Python API
    • Add v2 module aliases for:
    • tf.initializers => tf.keras.initializers
    • tf.losses => tf.keras.losses & tf.metrics => tf.keras.metrics
    • tf.optimizers => tf.keras.optimizers
    • Add tf.keras.layers.AbstractRNNCell as the preferred implementation of RNN cell for TF v2. User can use it to implement RNN cell with custom behavior.
    • Adding clear_losses API to be able to clear losses at the end of forward pass in a custom training loop in eager.
    • Add support for passing list of lists to the metrics param in Keras compile.
    • Added top-k to precision and recall to keras metrics.
    • Adding public APIs for cumsum and cumprod keras backend functions.
    • Fix: model.add_loss(symbolic_tensor) should work in ambient eager.
    • Add name argument to tf.string_split and tf.strings_split
    • Minor change to SavedModels exported from Keras using tf.keras.experimental.export. (SignatureDef key for evaluation mode is now "eval" instead of "test"). This will be reverted back to "test" in the near future.
    • Updates binary cross entropy logic in Keras when input is probabilities. Instead of converting probabilities to logits, we are using the cross entropy formula for probabilities.
    • Raw TensorFlow functions can now be used in conjunction with the Keras Functional API during model creation. This obviates the need for users to create Lambda layers in most cases when using the Functional API. Like Lambda layers, TensorFlow functions that result in Variable creation or assign ops are not supported.
    • Keras training and validation curves are shown on the same plot.
    • Introduce dynamic constructor argument in Layer and Model, which should be set to True when using imperative control flow in the call method.
    • Removing of dtype in the constructor of initializers and partition_info in call.
  • New ops and improved op functionality
    • Add OpKernels for some stateless maps
    • Add v2 APIs for AUCCurve and AUCSummationMethod enums. #tf-metrics-convergence
    • Add tf.math.nextafter op.
    • Add CompositeTensor base class.
    • Add tf.linalg.tridiagonal_solve op.
    • Add opkernel templates for common table operations.
    • Added support for TFLite in TensorFlow 2.0.
    • Adds summary trace API for collecting graph and profile information.
    • Add batch_dims argument to tf.gather.
    • Add support for add_metric in the graph function mode.
    • Add C++ Gradient for BatchMatMulV2.
    • Added tf.random.binomial
    • Added gradient for SparseToDense op.
    • Add legacy string flat hash map op kernels
    • Add a ragged size op and register it to the op dispatcher
    • Add broadcasting support to tf.matmul.
    • Add ellipsis (...) support for tf.einsum()
    • Added LinearOperator.adjoint and LinearOperator.H (alias).
    • Added GPU implementation of tf.linalg.tridiagonal_solve.
    • Added strings.byte_split
    • Add RaggedTensor.placeholder()
    • Add a new "result_type" parameter to tf.strings.split
    • add_update can now be passed a zero-arg callable in order to support turning off the update when setting trainable=False on a Layer of a Model compiled with run_eagerly=True.
    • Add variant wrapper for absl::string_view
    • Add expand_composites argument to all nest.* methods.
    • Add pfor converter for Squeeze.
    • Bug fix for tf.tile gradient
    • Expose CriticalSection in core as tf.CriticalSection.
    • Update Fingerprint64Map to use aliases
    • ResourceVariable support for gather_nd.
    • ResourceVariable's gather op supports batch dimensions.
    • Variadic reduce is supported on CPU
    • Extend tf.function with basic support for CompositeTensors arguments (such as SparseTensor and RaggedTensor).
    • Add templates and interfaces for creating lookup tables
    • Post-training quantization tool supports quantizing weights shared by multiple operations. The models made with versions of this tool will use INT8 types for weights and will only be executable interpreters from this version onwards.
    • Malformed gif images could result in an access out of bounds in the color palette of the frame. This has been fixed now
    • image.resize now considers proper pixel centers and has new kernels (incl. anti-aliasing).
    • Added an isotonic regression solver (tf.nn.isotonic_regression).
  • Performance
    • Turn on MKL-DNN contraction kernels by default. MKL-DNN dynamically dispatches the best kernel implementation based on CPU vector architecture. To disable them, build with --define=tensorflow_mkldnn_contraction_kernel=0.
    • Support for multi-host ncclAllReduce in Distribution Strategy.
    • Expose a flag that allows the number of threads to vary across Python benchmarks.
  • TensorFlow 2.0 Development
    • Add v2 sparse categorical crossentropy metric.
    • Allow non-Tensors through v2 losses.
    • Add UnifiedGRU as the new GRU implementation for tf2.0. Change the default recurrent activation function for GRU from 'hard_sigmoid' to 'sigmoid', and 'reset_after' to True in 2.0. Historically recurrent activation is 'hard_sigmoid' since it is fast than 'sigmoid'. With new unified backend between CPU and GPU mode, since the CuDNN kernel is using sigmoid, we change the default for CPU mode to sigmoid as well. With that, the default GRU will be compatible with both CPU and GPU kernel. This will enable user with GPU to use CuDNN kernel by default and get a 10x performance boost in training. Note that this is checkpoint breaking change. If user want to use their 1.x pre-trained checkpoint, please construct the layer with GRU(recurrent_activation='hard_sigmoid', reset_after=False) to fallback to 1.x behavior.
    • TF 2.0 - Update metric name to always reflect what the user has given in compile. Affects following cases 1. When name is given as 'accuracy'/'crossentropy' 2. When an aliased function name is used eg. 'mse' 3. Removing the weighted prefix from weighted metric names.
    • Begin adding Go wrapper for C Eager API
    • image.resize in 2.0 now supports gradients for the new resize kernels.
    • removed tf.string_split from v2 API
    • Expose tf.contrib.proto.* ops in tf.io (they will exist in TF2)
    • "Updates the TFLiteConverter API in 2.0. Changes from_concrete_function to from_concrete_functions."
    • Enable tf.distribute.experimental.MultiWorkerMirroredStrategy working in eager mode.
    • Support both binary and -1/1 label input in v2 hinge and squared hinge losses.
  • TensorFlow Lite
    • "Adds support for tflite_convert in 2.0."
    • "Remove lite.OpHint, lite.experimental, and lite.constant from 2.0 API."
  • tf.contrib
  • tf.data
    • Add num_parallel_reads and passing in a Dataset containing filenames into TextLineDataset and FixedLengthRecordDataset
    • Going forward we operate in TF 2.0, this change is part of the effort to slowly converting XYZDataset to DatasetV2 type which is the official version going to be used in TF 2.0 and motivated by some compatibility issue found, _BigtableXYZDataset (of type DatasetV2) does not implement the _as_variant_tensor() of DatasetV1, when moving contrib.bigtable to tensorflow_io. Converting into DatasetV2 removes the overheads to maintain V1 while we are moving into TF 2.0.
    • Add dataset ops to the graph (or create kernels in Eager execution) during the python Dataset object creation instead doing it during Iterator creation time.
    • Add support for TensorArrays to tf.data Dataset.
    • Switching tf.data functions to use defun, providing an escape hatch to continue using the legacy Defun.
  • Toolchains
    • CUDNN_INSTALL_PATH, TENSORRT_INSTALL_PATH, NCCL_INSTALL_PATH, NCCL_HDR_PATH are deprecated. Use TF_CUDA_PATHS instead which supports a comma-separated list of base paths that are searched to find CUDA libraries and headers.
    • TF code now resides in tensorflow_core and tensorflow is just a virtual pip package. No code changes are needed for projects using TensorFlow, the change is transparent
  • XLA
    • XLA HLO graphs can be inspected with interactive_graphviz tool now.
  • Estimator
    • Use tf.compat.v1.estimator.inputs instead of tf.estimator.inputs
    • Replace contrib references with tf.estimator.experimental.* for apis in early_stopping.py

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

1e100, 4d55397500, a6802739, abenmao, Adam Weiss, Ag Ramesh, Alan Du, Albin Joy, Alex, Aman Patel, Amit, Amit Kumar Jaiswal, Amit Srivastava, Andreas Eberle, Andy Craze, Anthony Platanios, Armen Poghosov, armenpoghosov, arp95, Arpit Shah, Ashwin Ramaswami, Aurelien Geron, AuréLien Geron, aweers, awesomealex1, Ayush Agrawal, Ben Barsdell, Bharat Raghunathan, Bhavani Subramanian, blairhan, BléNesi Attila, Brandon Carter, candy.dc, Chao Liu, chenchc, chie8842, Christian Hansen, Christian Sigg, Clayne Robison, crafet, csukuangfj, ctiijima, Dan Jarvis, Dan Lazewatsky, Daniel Ingram, Daniel Salvadori, Dave Airlie, David Norman, Dayananda V, Dayananda-V, delock, Denis Khalikov, Deven Desai, Dheeraj Rajaram Reddy, dmitrievanthony, Donovan Ong, Drew Szurko, Duncan Riach, Dustin Neighly, Edward Forgacs, EFanZh, Fei Hu, Felix Lemke, Filip Matzner, fo40225, frreiss, Gautam, gehring, Geoffrey Irving, Grzegorz George Pawelczak, Grzegorz Pawelczak, Gyoung-Yoon Ryoo, HanGuo97, Hanton Yang, Hari Shankar, hehongliang, Heungsub Lee, Hoeseong Kim, I-Hong Jhuo, Ilango R, Innovimax, Irene Dea, Jacky Ko, Jakub Lipinski, Jason Zaman, jcf94, Jeffrey Poznanovic, Jens Elofsson, Jeroen BéDorf, Jia Qingtong, Jiankang, Joe Q, Joe Quadrino, Joeran Beel, Jonas Rauber, Jonathan, Jonathan Kyl, Joppe Geluykens, Joseph Friedman, jtressle, jwu, K Yasaswi Sri Chandra Gandhi, K. Hodges, Kaixi Hou, Karl Lessard, Karl Weinmeister, Karthik Muthuraman, Kashif Rasul, KDR, Keno Fischer, Kevin Mader, kjopek, Koan-Sin Tan, kouml, ktaebum, Lakshay Tokas, Laurent Le Brun, Letian Kang, Li, Guizi, Loo Rong Jie, Lucas Hendren, Lukas Geiger, Luke Han, luxupu, Ma, Guokai, Mahmoud Abuzaina, Mandar Deshpande, manhyuk, Marco Gaido, Marek Drozdowski, Mark Collier, Mark Ryan, mars20, Mateusz Chudyk, Matt Conley, MattConley, mbhuiyan, mdfaijul, Melissa Grueter, Michael KäUfl, MickaëL Schoentgen, Miguel Morin, Mihail Salnikov, Mike Arpaia, Mike Holcomb, monklof, Moses Marin, Mshr-H, nammbash, Natalia Gimelshein, Nayana-Ibm, neargye, Neeraj Pradhan, Nehal J Wani, Nick, Niels Ole Salscheider, Niranjan Hasabnis, nlewycky, Nuka-137, Nutti, olicht, P Sudeepam, Palmer Lao, Pan Daoxin, Pariksheet Pinjari, Pavel Samolysov, PENGWA, Pooya Davoodi, R S Nikhil Krishna, Rohit Gupta, Roman Soldatow, rthadur, Ruizhe, Ryan Jiang, Samantha Andow, Sami Kama, Sana-Damani, Saurabh Deoras, sdamani, seanshpark, Sebastien Iooss, Serv-Inc, Shahzad Lone, Shashank Gupta, Shashi, shashvat, shashvatshahi1998, Siju, Siju Samuel, Snease-Abq, Spencer Schaber, sremedios, srinivasan.narayanamoorthy, Steve Lang, Steve Nesae, Sumesh Udayakumaran, Supriya Rao, Taylor Jakobson, Taylor Thornton, Ted Chang, ThisIsPIRI, Thomas Deegan, Thomas Hagebols, tianyapiaozi, Tim Zaman, tomguluson92, Tongxuan Liu, TungJerry, v1incent, Vagif, vcarpani, Vikram Tiwari, Vishwak Srinivasan, Vitor-Alves, wangsiyu, wateryzephyr, WeberXie, WeijieSun, Wen-Heng (Jack) Chung, wenxizhu, Will Battel, William D. Irons, wyzhao, Xin, Yasuhiro Matsumoto, ymodak, Yong Tang, Younes Khoudli, Yuan Lin, Yves-Noel Weweler, Zantares, zjjott, 卜居, 王振华 (Wang Zhenhua), 黄鑫

Release 1.12.3

Bug Fixes and Other Changes

  • Updates png_archive dependency to 1.6.37 to not be affected by CVE-2019-7317, CVE-2018-13785, and CVE-2018-14048.
  • Updates sqlite dependency to 3.28.0 to not be affected by CVE-2018-20506, CVE-2018-20346, and CVE-2018-20505.

Release 1.12.2

Bug Fixes and Other Changes

  • Fixes a potential security vulnerability where carefully crafted GIF images can produce a null pointer dereference during decoding.

Release 1.13.0

Major Features and Improvements

  • TensorFlow Lite has moved from contrib to core. This means that Python modules are under tf.lite and source code is now under tensorflow/lite rather than tensorflow/contrib/lite.
  • TensorFlow GPU binaries are now built against CUDA 10 and TensorRT 5.0.
  • Support for Python3.7 on all operating systems.
  • Moved NCCL to core.

Behavioral changes

  • Disallow conversion of python floating types to uint32/64 (matching behavior of other integer types) in tf.constant.
  • Make the gain argument of convolutional orthogonal initializers (convolutional_delta_orthogonal, convolutional_orthogonal_1D, convolutional_orthogonal_2D, convolutional_orthogonal_3D) have consistent behavior with the tf.initializers.orthogonal initializer, i.e. scale the output l2-norm by gain and NOT by sqrt(gain). (Note that these functions are currently in tf.contrib which is not guaranteed backward compatible).

Bug Fixes and Other Changes

  • Documentation
    • Update the doc with the details about the rounding mode used in quantize_and_dequantize_v2.
    • Clarify that tensorflow::port::InitMain() should be called before using the TensorFlow library. Programs failing to do this are not portable to all platforms.
  • Deprecations and Symbol renames.
    • Removing deprecations for the following endpoints: tf.acos, tf.acosh, tf.add, tf.as_string, tf.asin, tf.asinh, tf.atan, tf.atan2, tf.atanh, tf.cos, tf.cosh, tf.equal, tf.exp, tf.floor, tf.greater, tf.greater_equal, tf.less, tf.less_equal, tf.log, tf.logp1, tf.logical_and, tf.logical_not, tf.logical_or, tf.maximum, tf.minimum, tf.not_equal, tf.sin, tf.sinh, tf.tan
    • Deprecate tf.data.Dataset.shard.
    • Deprecate saved_model.loader.load which is replaced by saved_model.load and saved_model.main_op, which will be replaced by saved_model.main_op in V2.
    • Deprecate tf.QUANTIZED_DTYPES. The official new symbol is tf.dtypes.QUANTIZED_DTYPES.
    • Update sklearn imports for deprecated packages.
    • Deprecate Variable.count_up_to and tf.count_up_to in favor of Dataset.range.
    • Export confusion_matrix op as tf.math.confusion_matrix instead of tf.train.confusion_matrix.
    • Add tf.dtypes. endpoint for every constant in dtypes.py. Moving endpoints in versions.py to corresponding endpoints in tf.sysconfig. and tf.version.. Moving all constants under tf.saved_model submodules to tf.saved_model module. New endpoints are added in V1 and V2 but existing endpoint removals are only applied in V2.
    • Deprecates behavior where device assignment overrides collocation constraints inside a collocation context manager.
  • Keras & Python API
    • Add to Keras functionality analogous to tf.register_tensor_conversion_function.
    • Subclassed Keras models can now be saved through tf.contrib.saved_model.save_keras_model.
    • LinearOperator.matmul now returns a new LinearOperator.
  • New ops and improved op functionality
    • Add a Nearest Neighbor Resize op.
    • Add an ignore_unknown argument to parse_values which suppresses ValueError for unknown hyperparameter types. Such * Add tf.linalg.matvec convenience function.
    • tf.einsum()raises ValueError for unsupported equations like "ii->".
    • Add DCT-I and IDCT-I in tf.signal.dct and tf.signal.idct.
    • Add LU decomposition op.
    • Add quantile loss to gradient boosted trees in estimator.
    • Add round_mode to QuantizeAndDequantizeV2 op to select rounding algorithm.
    • Add unicode_encode, unicode_decode, unicode_decode_with_offsets, unicode_split, unicode_split_with_offset, and unicode_transcode ops. Amongst other things, this Op adds the ability to encode, decode, and transcode a variety of input text encoding formats into the main Unicode encodings (UTF-8, UTF-16-BE, UTF-32-BE)
    • Add "unit" attribute to the substr op, which allows obtaining the substring of a string containing unicode characters.
    • Broadcasting support for Ragged Tensors.
    • SpaceToDepth supports uint8 data type.
    • Support multi-label quantile regression in estimator.
    • We now use "div" as the default partition_strategy in tf.nn.safe_embedding_lookup_sparse, tf.nn.sampled_softmax and tf.nn.nce_loss. hyperparameter are ignored.
  • Performance
    • Improve performance of GPU cumsum/cumprod by up to 300x.
    • Added support for weight decay in most TPU embedding optimizers, including AdamW and MomentumW.
  • TensorFlow 2.0 Development
    • Add a command line tool to convert to TF2.0, tf_upgrade_v2
    • Merge tf.spectral into tf.signal for TensorFlow 2.0.
    • Change the default recurrent activation function for LSTM from 'hard_sigmoid' to 'sigmoid' in 2.0. Historically recurrent activation is 'hard_sigmoid' since it is fast than 'sigmoid'. With new unified backend between CPU and GPU mode, since the CuDNN kernel is using sigmoid, we change the default for CPU mode to sigmoid as well. With that, the default LSTM will be compatible with both CPU and GPU kernel. This will enable user with GPU to use CuDNN kernel by default and get a 10x performance boost in training. Note that this is checkpoint breaking change. If user want to use their 1.x pre-trained checkpoint, please construct the layer with LSTM(recurrent_activation='hard_sigmoid') to fallback to 1.x behavior.
  • TensorFlow Lite
    • Move from tensorflow/contrib/lite to tensorflow/lite.
    • Add experimental Java API for injecting TensorFlow Lite delegates
    • Add support for strings in TensorFlow Lite Java API.
  • tf.contrib:
    • Add Apache Ignite Filesystem plugin to support accessing Apache IGFS.
    • Dropout now takes rate argument, keep_prob is deprecated.
    • Estimator occurrences references tf.contrib.estimator were changed to tf.estimator:
    • tf.contrib.estimator.BaselineEstimator with tf.estimator.BaselineEstimator
    • tf.contrib.estimator.DNNLinearCombinedEstimator with tf.estimator.DNNLinearCombinedEstimator
    • tf.contrib.estimator.DNNEstimator with tf.estimator.DNNEstimator
    • tf.contrib.estimator.LinearEstimator with tf.estimator.LinearEstimator
    • tf.contrib.estimator.InMemoryEvaluatorHook and tf.estimator.experimental.InMemoryEvaluatorHook`.
    • tf.contrib.estimator.make_stop_at_checkpoint_step_hook with tf.estimator.experimental.make_stop_at_checkpoint_step_hook.
    • Expose `tf.distribute.Strategy as the new name for tf.contrib.distribute.DistributionStrategy.
    • Migrate linear optimizer from contrib to core.
    • Move tf.contrib.signal to tf.signal (preserving aliases in tf.contrib.signal).
    • Users of tf.contrib.estimator.export_all_saved_models and related should switch to tf.estimator.Estimator.experimental_export_all_saved_models.
  • tf.data:
    • Add tf.data.experimental.StatsOptions(), to configure options to collect statistics from tf.data.Dataset pipeline using StatsAggregator. Add nested option, experimental_stats (which takes a tf.data.experimen tal.StatsOptions object), to tf.data.Options. Deprecates tf.data.experimental.set_stats_agregator.
    • Performance optimizations:
    • Add tf.data.experimental.OptimizationOptions(), to configure options to enable tf.data performance optimizations. Add nested option, experimental_optimization (which takes a tf.data.experimental.OptimizationOptions object), to tf.data.Options. Remove performance optimization options from tf.data.Options, and add them under tf.data.experimental.OptimizationOptions instead.
    • Enable map_and_batch_fusion and noop_elimination optimizations by default. They can be disabled by configuring tf.data.experimental.OptimizationOptions to set map_and_batch = False or noop_elimination = False respectively. To disable all default optimizations, set apply_default_optimizations = False.
    • Support parallel map in map_and_filter_fusion.
    • Disable static optimizations for input pipelines that use non-resource tf.Variables.
    • Add NUMA-aware MapAndBatch dataset.
    • Deprecate tf.data.Dataset.make_one_shot_iterator() in V1, removed it from V2, and added tf.compat.v1.data.make_one_shot_iterator()`.
    • Deprecate tf.data.Dataset.make_initializable_iterator() in V1, removed it from V2, and added tf.compat.v1.data.make_initializable_iterator().
    • Enable nested dataset support in core tf.data transformations.
    • For tf.data.Dataset implementers: Added tf.data.Dataset._element_structured property to replace Dataset.output_{types,shapes,classes}.
    • Make num_parallel_calls of tf.data.Dataset.interleave and tf.data.Dataset.map work in Eager mode.
  • Toolchains
    • Fixed OpenSSL compatibility by avoiding EVP_MD_CTX_destroy.
    • Added bounds checking to printing deprecation warnings.
    • Upgraded CUDA dependency to 10.0
    • To build with Android NDK r14b, add "#include <linux/compiler.h>" to android-ndk-r14b/platforms/android-14/arch-*/usr/include/linux/futex.h
    • Removed :android_tensorflow_lib_selective_registration* targets, use :android_tensorflow_lib_lite* targets instead.
  • XLA
    • Move RoundToEven function to xla/client/lib/math.h.
    • A new environment variable TF_XLA_DEBUG_OPTIONS_PASSTHROUGH set to "1" or "true" allows the debug options passed within an XRTCompile op to be passed directly to the XLA compilation backend. If such variable is not set (service side), only a restricted set will be passed through.
    • Allow the XRTCompile op to return the ProgramShape resulted form the XLA compilation as a second return argument.
    • XLA HLO graphs can now be rendered as SVG/HTML.
  • Estimator
    • Replace all occurrences of tf.contrib.estimator.BaselineEstimator with tf.estimator.BaselineEstimator
    • Replace all occurrences of tf.contrib.estimator.DNNLinearCombinedEstimator with tf.estimator.DNNLinearCombinedEstimator
    • Replace all occurrences of tf.contrib.estimator.DNNEstimator with tf.estimator.DNNEstimator
    • Replace all occurrences of tf.contrib.estimator.LinearEstimator with tf.estimator.LinearEstimator
    • Users of tf.contrib.estimator.export_all_saved_models and related should switch to tf.estimator.Estimator.experimental_export_all_saved_models.
    • Update regression_head to the new Head API for Canned Estimator V2.
    • Switch multi_class_head to Head API for Canned Estimator V2.
    • Replace all occurrences of tf.contrib.estimator.InMemoryEvaluatorHook and tf.contrib.estimator.make_stop_at_checkpoint_step_hook with tf.estimator.experimental.InMemoryEvaluatorHook and tf.estimator.experimental.make_stop_at_checkpoint_step_hook
    • Migrate linear optimizer from contrib to core.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Abhinav Upadhyay, Ag Ramesh, akikaaa, Alexis Louis, Anders Huss, Andreas Madsen, Andrew Banchich, Andy Craze, Anton Dmitriev, Artem Malykh, Avijit-Nervana, Balint Cristian, Benjamin Tan Wei Hao, Bhavani Subramanian, Brendan Finan, Brian Nemsick, Bryan Cutler, By Shen, Cao Zongyan, Castiel, Chris Antaki, Christian Goll, Cibifang, Clayne Robison, Codrut Grosu, Cong Xu, Dalmo Cirne, Daniel Hunter, Dougal J. Sutherland, Edvard Fagerholm, EFanZh, Erik Smistad, Evgeniy Polyakov, Feiyang Chen, franklin5, Fred Reiss, Gautam, gehring, Geoffrey Irving, George Sterpu, Gitea, Grzegorz George Pawelczak, Guozhong Zhuang, himkt, Hoeseong Kim, Huan Li (李卓桓), HuiyangFei, hyunyoung, Isaac Burbank, jackonan, Jacky Ko, Jason Furmanek, Jason Zaman, Javier Luraschi, Jiang,Zhoulong, joaak, John Lin, Jonathan Wyatt Hoech, josephyearsley, Josh Gordon, Julian Niedermeier, Karl Lessard, Keno Fischer, lanhin, Leon Graser, leondgarse, Li, Guizi, Li, Yiqiang, lxl910915, Mahmoud Abuzaina, manhyuk, Marcela Morales Quispe, margaretmz, Matt Conley, Max Pumperla, mbhuiyan, mdfaijul, Meng, Peng, Michael, Michael Gielda, mrTsjolder, Muhammad Wildan, neargye, Nehal J Wani, NEWPLAN, Niranjan Hasabnis, Nutti, olicht, Pan Daoxin, Pedro Monreal, Peng Yu, pillarpond, Pooya Davoodi, qiezi, Rholais Lii, Richard Yu, Rin Arakaki, Roger Iyengar, sahilbadyal, Sami Kama, Sandip Giri, Scott Leishman, Serge Panev, Seunghoon Park, Shafi Dayatar, shengfuintel, Shimin Guo, Siju, silent567, Stefan Dyulgerov, steven, Tao Wei, Thor Johnsen, Tingbo Lu, tomguluson92, Tongxuan Liu, Trevor Morris, Ubuntu, Vadim Borisov, vanderliang, wangsiyu, Wen Yun, Wen-Heng (Jack) Chung, wenxizhu, William D. Irons, Xiaoming (Jason) Cui, Yan Facai (颜发才), Yanbo Liang, Yaniv Blumenfeld, Yash Gaurkar, Yicheng Fan, Yong Tang, Yongjoon Lee, Yuan (Terry) Tang, Yuxin Wu, zldrobit

Release 1.12.0

Major Features and Improvements

  • Keras models can now be directly exported to the SavedModel format(tf.contrib.saved_model.save_keras_model()) and used with Tensorflow Serving.
  • Keras models now support evaluating with a tf.data.Dataset.
  • TensorFlow binaries are built with XLA support linked in by default.
  • Ignite Dataset added to contrib/ignite that allows to work with Apache Ignite.

Bug Fixes and Other Changes

  • tf.data:
    • tf.data users can now represent, get, and set options of TensorFlow input pipelines using tf.data.Options(), tf.data.Dataset.options(), and tf.data.Dataset.with_options() respectively.
    • New tf.data.Dataset.reduce() API allows users to reduce a finite dataset to a single element using a user-provided reduce function.
    • New tf.data.Dataset.window() API allows users to create finite windows of input dataset; when combined with the tf.data.Dataset.reduce() API, this allows users to implement customized batching.
    • All C++ code moves to the tensorflow::data namespace.
    • Add support for num_parallel_calls to tf.data.Dataset.interleave.
  • tf.contrib:
    • Remove tf.contrib.linalg. tf.linalg should be used instead.
    • Replace any calls to tf.contrib.get_signature_def_by_key(metagraph_def, signature_def_key) with meta_graph_def.signature_def[signature_def_key]. Catching a ValueError exception thrown by tf.contrib.get_signature_def_by_key should be replaced by catching a KeyError exception.
  • tf.contrib.data
    • Deprecate, and replace by tf.data.experimental.
  • Other:
    • Instead of jemalloc, revert back to using system malloc since it simplifies build and has comparable performance.
    • Remove integer types from tf.nn.softplus and tf.nn.softsign OpDefs. This is a bugfix; these ops were never meant to support integers.
    • Allow subslicing Tensors with a single dimension.
    • Add option to calculate string length in Unicode characters.
    • Add functionality to SubSlice a tensor.
    • Add searchsorted (ie lower/upper_bound) op.
    • Add model explainability to Boosted Trees.
    • Support negative positions for tf.substr.
    • There was previously a bug in the bijector_impl where the _reduce_jacobian_det_over_event does not handle scalar ILDJ implementations properly.
    • In tf eager execution, allow re-entering a GradientTape context.
    • Add tf_api_version flag. If --define=tf_api_version=2 flag is passed in, then bazel will build TensorFlow API version 2.0. Note that TensorFlow 2.0 is under active development and has no guarantees at this point.
    • Add additional compression options to TfRecordWriter.
    • Performance improvements for regex full match operations.
    • Replace tf.GraphKeys.VARIABLES with tf.GraphKeys.GLOBAL_VARIABLES.
    • Remove unused dynamic learning rate support.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

(David) Siu-Kei Muk, Ag Ramesh, Anton Dmitriev, Artem Sobolev, Avijit-Nervana, Bairen Yi, Bruno Goncalves, By Shen, candy.dc, Cheng Chen, Clayne Robison, coder3101, Dao Zhang, Elms, Fei Hu, feiquan, Geoffrey Irving, Guozhong Zhuang, hellcom, Hoeseong Kim, imsheridan, Jason Furmanek, Jason Zaman, Jenny Sahng, jiefangxuanyan, Johannes Bannhofer, Jonathan Homer, Koan-Sin Tan, kouml, Loo Rong Jie, Lukas Geiger, manipopopo, Ming Li, Moritz KröGer, Naurril, Niranjan Hasabnis, Pan Daoxin, Peng Yu, pengwa, rasmi, Roger Xin, Roland Fernandez, Sami Kama, Samuel Matzek, Sangjung Woo, Sergei Lebedev, Sergii Khomenko, shaohua, Shaohua Zhang, Shujian2015, Sunitha Kambhampati, tomguluson92, ViníCius Camargo, wangsiyu, weidankong, Wen-Heng (Jack) Chung, William D. Irons, Xin Jin, Yan Facai (颜发才), Yanbo Liang, Yash Katariya, Yong Tang, 在原佐为

Release 1.11.0

Major Features and Improvements

  • Nvidia GPU:
  • Google Cloud TPU:
    • Experimental tf.data integration for Keras on Google Cloud TPUs.
    • Experimental / preview support for eager execution on Google Cloud TPUs.
  • DistributionStrategy:
    • Add multi-GPU DistributionStrategy support in tf.keras. Users can now use fit, evaluate and predict to distribute their model on multiple GPUs.
    • Add multi-worker DistributionStrategy and standalone client support in Estimator. See README for more details.
  • Add C, C++, and Python functions for querying kernels.

Breaking Changes

  • Keras:
    • The default values for tf.keras RandomUniform, RandomNormal, and TruncatedNormal initializers have been changed to match those in external Keras.
    • Breaking change: model.get_config() on a Sequential model now returns a config dictionary (consistent with other Model instances) instead of a list of configs for the underlying layers.

Bug Fixes and Other Changes

  • C++:
    • Changed the signature of SessionFactory::NewSession so that it can return a meaningful error message on failure.
  • tf.data:
    • Remove num_parallel_parser_calls argument from tf.contrib.data.make_csv_dataset(). [tf.data] Remove num_parallel_parser_calls argument from tf.contrib.data.make_csv_dataset().
    • tf.data.Dataset.list_files() raises an exception at initialization time if the argument matches no files.
    • Renamed BigTable class to BigtableTable for clarity
    • Document use of the Cloud Bigtable API
    • Add tf.contrib.data.reduce_dataset which can be used to reduce a dataset to a single element.
    • Generalization of tf.contrib.data.sliding_window_batch.
  • INC:
    • Runtime improvements to triangular solve.
  • tf.contrib:
    • Add an implementation argument to tf.keras.layers.LocallyConnected2D and tf.keras.layers.LocallyConnected1D. The new mode (implementation=2) performs forward pass as a single dense matrix multiplication, allowing dramatic speedups in certain scenarios (but worse performance in others - see docstring). The option also allows to use padding=same.
    • Add documentation clarifying the differences between tf.fill and tf.constant.
    • Add experimental IndexedDatasets.
    • Add selective registration target using the lite proto runtime.
    • Add simple Tensor and DataType classes to TensorFlow Lite Java
    • Add support for bitcasting to/from uint32 and uint64.
    • Added a subclass of Estimator that can be created from a SavedModel (SavedModelEstimator).
    • Adds leaf index modes as an argument.
    • Allow a different output shape from the input in tf.contrib.image.transform.
    • Change the state_size order of the StackedRNNCell to be natural order. To keep the existing behavior, user can add reverse_state_order=True when constructing the StackedRNNCells.
    • Deprecate self.test_session() in favor of self.session() or self.cached_session().
    • Directly import tensor.proto.h (the transitive import will be removed from tensor.h soon).
    • Estimator.train() now supports tf.contrib.summary.* summaries out of the box; each call to .train() will now create a separate tfevents file rather than re-using a shared one.
    • Fix FTRL L2-shrinkage behavior: the gradient from the L2 shrinkage term should not end up in the accumulator.
    • Fix toco compilation/execution on Windows.
    • GoogleZoneProvider class added to detect which Google Cloud Engine zone tensorflow is running in.
    • It is now safe to call any of the C API's TF_Delete* functions on nullptr.
    • Log some errors on Android to logcat.
    • Match FakeQuant numerics in TFLite to improve accuracy of TFLite quantized inference models.
    • Optional bucket location check for the GCS Filesystem.
    • Performance enhancements for StringSplitOp & StringSplitV2Op.
    • Performance improvements for regex replace operations.
    • TFRecordWriter now raises an error if .write() fails.
    • TPU: More helpful error messages in TPUClusterResolvers.
    • The legacy_init_op argument to SavedModelBuilder methods for adding MetaGraphs has been deprecated. Please use the equivalent main_op argument instead. As part of this, we now explicitly check for a single main_op or legacy_init_op at the time of SavedModel building, whereas the check on main_op was previously only done at load time.
    • The protocol used for Estimator training is now configurable in RunConfig.
    • Triangular solve performance improvements.
    • Unify RNN cell interface between TF and Keras. Add new get_initial_state() to Keras and TF RNN cell, which will use to replace the existing zero_state() method.
    • Update initialization of variables in Keras.
    • Updates to "constrained_optimization" in tensorflow/contrib.
    • boosted trees: adding pruning mode.
    • tf.train.Checkpoint does not delete old checkpoints by default.
    • tfdbg: Limit the total disk space occupied by dumped tensor data to 100 GBytes. Add environment variable TFDBG_DISK_BYTES_LIMIT to allow adjustment of this upper limit.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Aapeli, adoda, Ag Ramesh, Amogh Mannekote, Andrew Gibiansky, Andy Craze, Anirudh Koul, Aurelien Geron, Avijit, Avijit-Nervana, Ben, Benjamin H. Myara, bhack, Brett Koonce, Cao Zongyan, cbockman, cheerss, Chikanaga Tomoyuki, Clayne Robison, cosine0, Cui Wei, Dan J, David, David Norman, Dmitry Klimenkov, Eliel Hojman, Florian Courtial, fo40225, formath, Geoffrey Irving, gracehoney, Grzegorz Pawelczak, Guoliang Hua, Guozhong Zhuang, Herman Zvonimir DošIlović, HuiyangFei, Jacker, Jan HüNnemeyer, Jason Taylor, Jason Zaman, Jesse, Jiang,Zhoulong, Jiawei Zhang, Jie, Joe Yearsley, Johannes Schmitz, Jon Perl, Jon Triebenbach, Jonathan, Jonathan Hseu, Jongmin Park, Justin Shenk, karl@kubx.ca, Kate Hodesdon, Kb Sriram, Keishi Hattori, Kenneth Blomqvist, Koan-Sin Tan, Li Liangbin, Li, Yiqiang, Loo Rong Jie, Madiyar, Mahmoud Abuzaina, Mark Ryan, Matt Dodge, mbhuiyan, melvinljy96, Miguel Mota, Nafis Sadat, Nathan Luehr, naurril, Nehal J Wani, Niall Moran, Niranjan Hasabnis, Nishidha Panpaliya, npow, olicht, Pei Zhang, Peng Wang (Simpeng), Peng Yu, Philipp Jund, Pradeep Banavara, Pratik Kalshetti, qwertWZ, Rakesh Chada, Randy West, Ray Kim, Rholais Lii, Robin Richtsfeld, Rodrigo Silveira, Ruizhi, Santosh Kumar, Seb Bro, Sergei Lebedev, sfujiwara, Shaba Abhiram, Shashi, SneakyFish5, Soila Kavulya, Stefan Dyulgerov, Steven Winston, Sunitha Kambhampati, Surry Shome, Taehoon Lee, Thor Johnsen, Tristan Rice, TShapinsky, tucan, tucan9389, Vicente Reyes, Vilmar-Hillow, Vitaly Lavrukhin, wangershi, weidan.kong, weidankong, Wen-Heng (Jack) Chung, William D. Irons, Wim Glenn, XFeiF, Yan Facai (颜发才), Yanbo Liang, Yong Tang, Yoshihiro Yamazaki, Yuan (Terry) Tang, Yuan, Man, zhaoyongke, ÁRon Ricardo Perez-Lopez, 张天启, 张晓飞

Release 1.10.1

Bug Fixes and Other Changes

  • tf.keras:
    • Fixing keras on Cloud TPUs. No new binaries will be built for Windows.

Release 1.10.0

Major Features And Improvements

  • The tf.lite runtime now supports complex64.
  • Initial Google Cloud Bigtable integration for tf.data.
  • Improved local run behavior in tf.estimator.train_and_evaluate which does not reload checkpoints for evaluation.
  • RunConfig now sets device_filters to restrict how workers and PS can communicate. This can speed up training and ensure clean shutdowns in some situations. But if you have jobs that require communication between workers, you will have to set custom session_options in your RunConfig.
  • Moved Distributions and Bijectors from tf.contrib.distributions to Tensorflow Probability (TFP). tf.contrib.distributions is now deprecated and will be removed by the end of 2018.
  • Adding new endpoints for existing tensorflow symbols. These endpoints are going to be the preferred endpoints going forward and may replace some of the existing endpoints in the future. See below for the complete list. New symbols have been added to the following modules: tf.debugging, tf.dtypes, tf.image, tf.io, tf.linalg, tf.manip, tf.math, tf.quantization, tf.strings

Breaking Changes

  • Prebuilt binaries are now (as of TensorFlow 1.10) built against NCCL 2.2 and no longer include NCCL in the binary install. TensorFlow usage with multiple GPUs and NCCL requires upgrade to NCCL 2.2. See updated install guides: TensorFlow GPU support and Build TensorFlow from source.
  • Starting from TensorFlow 1.11, Windows builds will use Bazel. Therefore, we will drop official support for cmake.

Bug Fixes and Other Changes

  • tf.data:
    • tf.contrib.data.group_by_reducer() is now available via the public API.
    • tf.contrib.data.choose_from_datasets() is now available via the public API.
    • Adding drop_remainder argument to tf.data.Dataset.batch() and tf.data.Dataset.padded_batch(), deprecating tf.contrib.data.batch_and_drop_remainder() and tf.contrib.data.padded_batch_and_drop_remainder().
  • tf.estimator:
    • Estimators now use custom savers included in EstimatorSpec scaffolds for saving SavedModels during export.
    • EstimatorSpec will now add a default prediction output for export if no export_output is provided, eliminating the need to explicitly include a PredictOutput object in the model_fn for simple use-cases.
    • Support sparse_combiner in canned Linear Estimators.
    • Added batch normalization to DNNClassifier, DNNRegressor, and DNNEstimator.
    • Adding ranking support for boosted trees.
    • Adding center bias option for boosted trees.
  • Add synchronization and aggregation args to get_variable(). These args will be used for distributed variables.
  • Add synchronization and aggregation args to the layer add_weight() API. These args will be used for distributed variables.
  • tf.losses.* do not add to the global collection when executing eagerly (to avoid leaking memory).
  • Support different summary and checkpoint directories in tf.train.MonitoredTrainingSession().
  • Added IndRNN, IndyGRU, and IndyLSTM cells to tf.contrib.rnn.
  • Add safe static factory functions for SparseTensor and convert all CHECKs to DCHECKs. Using the constructor directly is unsafe and deprecated.
  • Make the Bigtable client connection pool configurable & increase the default # of connections for performance.
  • Added derivative of tf.random_gamma with respect to the alpha parameter.
  • Added derivative of tf.igamma(a, x) and tf.igammac(a, x) with respect to a.
  • Modified Bessel functions of order zero and one.
  • Add FillTriangular Bijector to create triangular matrices.
  • Added support for Type III DCT, and tf.spectral.idct(type=2|3).
  • Correctly handle CuDNN RNN weight loaded when nest in TimeDistributed.
  • Adding per-element weight support for WALSComputePartialLhsAndRhsOp.
  • ZerosLike and OnesLike ops treated as constants by Graph Transform Tool.
  • Gamma distribution and the derived distributions (Beta, Dirichlet, Student's t, inverse Gamma) now fully reparameterized.
  • Java: Experimental wrapper classes to make graph generation easier. Thanks @karllessard and @kbsriram
  • Build & link in secure gRPC components (switch from the insecure grpc dependency to secure grpc dependency).
  • Adding new endpoints for existing tensorflow symbols. These endpoints are going to be the preferred endpoints going forward and may replace some of the existing endpoints in the future. List of new endpoints:
    • New endpoints in tf.image namespace: tf.image.extract_image_patches
    • New endpoints in tf.debugging namespace: tf.debugging.check_numerics, tf.debugging.is_finite, tf.debugging.is_inf, tf.debugging.is_nan.
    • New endpoints in tf.dtypes namespace: tf.dtypes.as_string.
    • New endpoints in tf.io namespace: tf.io.decode_base64, tf.io.decode_compressed, tf.io.decode_json_example, tf.io.decode_raw, tf.io.encode_base64, tf.io.matching_files, tf.io.parse_tensor, tf.io.read_file,tf.io.write_file`.
    • New endpoints in tf.linalg namespace: tf.linalg.cross, tf.linalg.tensor_diag (corresponds to tf.diag), tf.linalg.tensor_diag_part (corresponds to tf.diag_part).
    • New endpoints in tf.manip namespace: tf.manip.batch_to_space_nd, tf.manip.gather_nd, tf.manip.reshape, tf.manip.reverse, tf.manip.scatter_nd, tf.manip.space_to_batch_nd, tf.manip.tile
    • New endpoints in tf.math namespace: tf.math.acos, tf.math.acosh, tf.math.add, tf.math.asin, tf.math.asinh, tf.math.atan, tf.math.atan2, tf.math.atanh, tf.math.betainc, tf.math.ceil, tf.math.cos, tf.math.cosh, tf.math.digamma, tf.math.equal, tf.math.erfc, tf.math.exp, tf.math.expm1, tf.math.floor, tf.math.greater, tf.math.greater_equal, tf.math.igamma, tf.math.igammac, tf.math.invert_permutation, tf.math.less, tf.math.less_equal, tf.math.lgamma, tf.math.log, tf.math.log1p, tf.math.logical_and, tf.math.logical_not, tf.math.logical_or, tf.math.maximum, tf.math.minimum, tf.math.not_equal, tf.math.polygamma, tf.math.reciprocal, tf.math.rint, tf.math.rsqrt, tf.math.segment_max, tf.math.segment_mean, tf.math.segment_min, tf.math.segment_prod, tf.math.segment_sum, tf.math.sin, tf.math.sinh, tf.math.softplus, tf.math.softsign, tf.math.squared_difference, tf.math.tan, tf.math.unsorted_segment_max, tf.math.unsorted_segment_min, tf.math.unsorted_segment_prod, tf.math.unsorted_segment_sum, tf.math.zeta.
    • New endpoints in tf.quantization namespace: tf.quantization.dequantize, tf.quantization.fake_quant_with_min_max_args, tf.quantization.fake_quant_with_min_max_args_gradient, tf.quantization.fake_quant_with_min_max_vars, tf.quantization.fake_quant_with_min_max_vars_gradient, tf.quantization.fake_quant_with_min_max_vars_per_channel, tf.quantization.fake_quant_with_min_max_vars_per_channel_gradient.
    • New endpoints in tf.strings namespace: tf.strings.join (corresponds to tf.string_join), tf.strings.regex_replace, tf.strings.to_number (corresponds to tf.string_to_number), tf.strings.strip (corresponds to tf.string_strip), tf.strings.substr, tf.strings.to_hash_bucket (corresponds to tf.string_to_hash_bucket), tf.strings.to_hash_bucket_fast (corresponds to tf.string_to_hash_bucket_fast), tf.strings.to_hash_bucket_strong (corresponds to tf.string_to_hash_bucket_strong).

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Ag Ramesh, Alex Wiltschko, Alexander Pantyukhin, Amogh Mannekote, An Jiaoyang, Andrei Nigmatulin, Andrew Ginns, BjøRn Moholt, Brett Koonce, Chengzhi Chen, Chinmay Das, Christian Ertler, Christoph Boeddeker, Clayne Robison, Courtial Florian, ctiijima, Dan Douthit, Dan J, Dan Ringwalt, EFanZh, Emanuele Ballarin, eqy, Evgeniy Zheltonozhskiy, Freedom" Koan-Sin Tan, FréDéRic Branchaud-Charron, G K, gracehoney, Guillaume Klein, Guozhong Zhuang, Hsien-Yang Li, hsm207, ImSheridan, Jayaram Bobba, Jiandong Ruan, Jie, Joel Shor, Jonas Rauber, Jongmin Baek, jsawruk, Karan Kaw, Karl Lessard, karl@kubx.ca, Kb Sriram, KinmanLam, leiiwang, Li, Yiqiang, Loo Rong Jie, Mahmoud Abuzaina, Mahmoud Aslan, ManHyuk, Martin Patz, Martin Zeitler, mktozk, Mohammad Ashraf Bhuiyan, mrTsjolder, Naman Bhalla, Nick Felt, Nicolas Lopez, Niranjan Hasabnis, Nishidha Panpaliya, Nitish, nrstott, Nutti, Parag Jain, PeterLee, Philipp Jund, Rach L, Rafal Wojdyla, Roland Zimmermann, Sergei Lebedev, SneakyFish5, Soila Kavulya, Sriram Veturi, Steven Schmatz, Taehoon Lee, Tang, Wenyi, Taras Sereda, Ted Chang, Tim Zaman, Tristan Rice, tucan, vchigrin, Vikram Tiwari, Vincent, WeberXie, William D. Irons, Yan Facai (颜发才), Yong Tang, Yu Yi, Yuxin Wu, Zé ViníCius

Release 1.9.0

Major Features And Improvements

Breaking Changes

  • If you're opening empty variable scopes; replace variable_scope('', ...) by variable_scope(tf.get_variable_scope(), ...).
  • Headers used for building custom ops have been moved from site-packages/external into site-packages/tensorflow/include/external.

Bug Fixes and Other Changes

  • tfe.Network is deprecated. Please inherit from tf.keras.Model.
  • Layered variable names have changed in the following conditions:
    • Using tf.keras.layers with custom variable scopes.
    • Using tf.layers in a subclassed tf.keras.Model class. See here for more details
  • tf.data:
    • Dataset.from_generator() now accepts an args list, in order to create nested generators.
    • Dataset.list_files() now produces deterministic results when shuffle=False or a seed is passed.
    • tf.contrib.data.sample_from_datasets() and tf.contrib.data.choose_from_datasets() make it easier to sample or deterministically choose elements from multiple datasets.
    • tf.contrib.data.make_csv_dataset() now supports line breaks in quoted strings, and two infrequently used arguments removed.
    • (C++) DatasetBase::DebugString() is now const.
    • (C++) DatasetBase::MakeIterator() has been renamed to DatasetBase::MakeIteratorInternal().
    • (C++) IteratorBase::Initialize() method was added to support raising errors during iterator construction.
  • Eager Execution:
    • Added the ability to pause recording operations for gradient computation via tf.GradientTape.stop_recording.
    • Updated documentation, introductory notebooks.
  • tf.keras:
    • Move Keras code out of _impl folder and remove API files.
    • tf.keras.Model.save_weights now saves in TensorFlow format by default.
    • Enable dataset iterators to be passed to tf.keras.Model training/eval methods.
  • TensorFlow Debugger (tfdbg) CLI: fix an issue in which the TensorBoard Debugger Plugin could not handle total source file size exceeding gRPC message size limit (4 MB).
  • tf.contrib:
    • tf.contrib.framework.zero_initializer supports ResourceVariable.
    • Adding "constrained_optimization" to tensorflow/contrib.
  • Other:
    • Add GCS Configuration Ops.
    • Changing signature of MakeIterator to enable propagating error status.
    • KL divergence for two Dirichlet distributions.
    • More consistent GcsFileSystem behavior for certain reads past EOF.
    • Update benchmark for tf.scan to match ranges across eager and graph modes.
    • Fixed bug in tf.reduce_prod gradient for complex dtypes.
    • Allow the use of '.' in variables (e.g. "hparams.parse('a.b=1.0')"), which would previously raise an error. This will correspond to an attribute name with an embedded '.' symbol (e.g. 'a.b'), which can only be accessed indirectly (e.g. through getattr and setattr). To set this up the user will first need to explicitly add the variable to the hparam object (e.g. "hparams.add_hparam(name='a.b', value=0.0)").
    • Benchmark for tf.scan in graph and eager modes.
    • Added complex128 support to FFT, FFT2D, FFT3D, IFFT, IFFT2D, and IFFT3D.
    • Making ids unique in nn.embedding_lookup_sparse. This helps to reduce RPC calls for looking up the embeddings when there are repeated ids in the batch.
    • Support indicator column in boosted trees.
    • Prevent tf.gradients() from backpropagating through integer tensors.
    • LinearOperator[1D,2D,3D]Circulant added to tensorflow.linalg.
    • Conv3D, Conv3DBackpropInput, Conv3DBackpropFilter now supports arbitrary.
    • Added tf.train.Checkpoint for reading/writing object-based checkpoints.
    • Added LinearOperatorKronecker, a dense-free implementation of the Kronecker Product.
    • Allow LinearOperator to broadcast.
    • SavedModelBuilder will now deduplicate asset names that point to files with the same basename and the same contents. Note that this may result in new asset files included in SavedModels in cases where assets with the same name but different contents were previously overwriting each other.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

Abdullah Alrasheed, Achal Shah, Ad-530, ADiegoCAlonso, Aditya Yogi, Ag Ramesh, akindyakov, Andy Kernahan, Anya Petrova, Aurelien Geron, Ben, Ben Barsdell, Bhavani-Subramanian, braincodercn, Brett Koonce, Brian Nemsick, Brian Zier, Bryan Heden, candy.dc, cclauss, Clayne Robison, ctiijima, Dalmo Cirne, David Norman, David T.H. Kao, DosLin, ekelsen, Elson Rodriguez, Erik Smistad, Felix Abecassis, Fergal Cotter, fo40225, foo0x29a, Freedom" Koan-Sin Tan, FréDéRic Branchaud-Charron, gdh1995, Geoffrey Irving, Giuseppe, gracehoney, Guido Zuidhof, Guillaume Klein, Guozhong Zhuang, Haggai, Harald Husum, imsheridan, Ivan Zhang, Jan Zikes, Jayaram Bobba, Jesse Benson, Jesse Gumz, Jiajia Li, Jie, jinghuangintel, Jingwen, jjsjann123, Joe Yearsley, Joel Hestness, Joel Shor, josephyearsley, Junpeng Lao, Karol M. Langner, Kb Sriram, krantideep95, Krish Ravindranath, Letian Feng, Loo Rong Jie, Lukas Geiger, Maciej, Mahmoud Abuzaina, ManHyuk, Mark Ryan, mbhuiyan, Michal Turek, Mostafa Alaa, Myungsung Kwak, Nand Dalal, Nehal J Wani, Neil Tenenholtz, ngc92, Nicholas Nadeau, P.Eng., Avs, Niranjan Hasabnis, P-Hidringer, Paul Van Eck, Peng Yu, Qing Zhao, Qingying Chen, Quanlong, Rajendra Arora, Rholais Lii, rmanyari, Robin Richtsfeld, Russell Klopfer, Sagi, Sam Sendelbach, Sandeep N Gupta, Sandip Giri, Sarah Edkins, Scott Tseng, Sdalbsoo, Sergii Khomenko, Seungwoo Choi (Biggie), Seyed Majid Azimi, Shaoning Zeng, shengfuintel, Siu Kei, Muk, Smit Shilu, soonson, Stefan Schweter, Sukhwan Kim, Sunitha Kambhampati, Taehoon Lee, tamimaddari82, Tang, Wenyi, Ted Chang, u2takey, Utkarsh Upadhyay, Vadim Markovtsev, voegtlel, Wai Hon Law, wangsiyu, Wenhao Hu, wenhao.hu, William D. Irons, Yan Facai (颜发才), Yanbo Liang, Yihong Wang, Yilei (Dolee) Yang, Yong Tang, Yuan (Terry) Tang

Release 1.8.0

Major Features And Improvements

  • Can now pass tf.contrib.distribute.MirroredStrategy() to tf.estimator.RunConfig() to run an Estimator model on multiple GPUs on one machine.
  • Add tf.contrib.data.prefetch_to_device(), which supports prefetching to GPU memory.
  • Added Gradient Boosted Trees as pre-made Estimators: BoostedTreesClassifier, BoostedTreesRegressor.
  • Add 3rd generation pipeline config for Cloud TPUs which improves performance and usability.
  • tf.contrib.bayesflow is moving out to it's own repo.
  • Added tf.contrib.{proto,rpc} to allow generic proto parsing and RPC communication1.

Bug Fixes and Other Changes

  • tf.data:
    • Add tf.contrib.data.prefetch_to_device, which enables prefetching dataset elements to GPU memory.
    • Add tf.contrib.data.AUTOTUNE, which allows the tf.data runtime to automatically tune the prefetch buffer sizes based on your system and environment.
    • Add tf.contrib.data.make_csv_dataset for building datasets of CSV files.
  • Eager Execution:
    • With eager execution Datasets can now be used as standard python iterators (for batch in dataset:). Both Dataset.__iter__() and Dataset.make_one_shot_iterator() can now be used to create iterators when eager execution is enabled.
    • Automatic device placement has been enabled (i.e., use a GPU if available automatically, without requiring an explicit with tf.device(“/gpu:0”)) (Fixes #14133)
    • tf.GradientTape has moved out of contrib.
  • tf.keras:
    • Added the fashion mnist dataset.
    • New data preprocessing functions: image/random_brightness, sequence/TimeseriesGenerator, and text/hashing_trick.
  • Accelerated Linear Algebra (XLA):
    • Select and scatter in reference util and evaluator now use lexicographical order to break ties.
  • TensorFlow Debugger (tfdbg) CLI:
    • During tensor-filter operations, allow exclusion of nodes by regular expressions.
    • Fix spurious background colors in some text terminals.
  • tf.contrib:
    • Add meta-distribution BatchReshape which reshapes batch dimensions.
    • tf.contrib.layers.recompute_grad works for explicit gradient checkpointing on TPU.
    • Add tf.contrib.framework.argsort.
    • Allow DNNBoostedTreeCombinedEstimator to work with core versions of feature columns and losses.
    • Add non-linear image warping ops: tf.contrib.image.sparse_image_warp, tf.contrib.image.dense_image_warp, and tf.contrib.image.interpolate_spline.
    • Fix bug in tf.contrib.opt.MultitaskOptimizerWrapper where types of tensors were mismatched.
  • Other:
    • Low-level graph construction now calls the TensorFlow C API. This change should be invisible to most users, but can be disabled by setting the environment variable TF_C_API_GRAPH_CONSTRUCTION=0 in this release. Future releases will remove the ability to disable this change. Please file a bug if you find yourself using this escape hatch.
    • Add description of shapes and a pointer to tutorial notebook in tf.distributions.Distribution.
    • Update scatter operations:
    • Add tf.scatter_min and tf.scatter_max
    • Extend scatter operations to work with a scalar update parameter.
    • Move cuDNN RNN ops to core for use in TensorFlow codebase only.
    • Add float64 support for Conv2d, Conv2dBackpropInput, and Conv2dBackpropFilter.
    • Add float64 support for AvgPool/AvgPoolGrad.
    • Make graph name scope thread local so that they work correctly in multi-threaded environments.
    • Update nsync synchronization library to avoid slow primitives on Linux.
    • Removed need to put nsync/public on C include path when building custom ops.
    • Add tf.image.psnr, tf.image.ssim, tf.image.ssim_multiscale, tf.image.image_gradients, tf.image.sobel_edges.
    • Add links to https://js.tensorflow.org.
    • Fix non-uniformity of orthogonal matrices.
    • Fix bug where multi-image Estimator eval summaries were not displayed correctly.

1 The cancellation logic of the RPC op contains a concurrency error. A fix has been submitted to master and will be part of the next release.

Thanks to our Contributors

This release contains contributions from many people at Google, as well as:

4d55397500, Aghasy, Alan Du, Alan Lee, Alan Yee, Alex Wiltschko, Animesh Karnewar, Ankit Gupta, Anton Matosov, Aris L, Ben Barsdell, Brent Yi, Brett Koonce, Carl Thomé, cbockman, Chikanaga Tomoyuki, Chris Tava, CéDric Deltheil, Dahan Gong, Dalmo Cirne, Daniel Erenrich, David Norman, DavidNorman, Edd Wilder-James, Fanjin Zeng, Felix Abecassis, fo40225, George Sterpu, Giovanni Terlingen, Gor Baghdasaryan, Guillaume Klein, Hanchen Li, Ilya Polenov, Jakub Kolodziejczyk, Jason Sadler, Jayaram Bobba, Jerry Liu, jinghuangintel, Jiongyan Zhang (张炯衍), Joel Shor, Jong Wook Kim, Julian Eisenschlos, Karl Lessard, Krish Ravindranath, Loo Rong Jie, Lukas Geiger, Luke Iwanski, Mahmoud Abuzaina, ManHyuk, Marvin Richter, Maximilian Mitchell, Mohammad Ashraf Bhuiyan, msofka, Mustafa Kasap, Nathan Burnham, Nathan Luehr, Naveen Marri, ngc92, nio1814, Oleg Zabluda, Ou Changkun, Panos Ipeirotis, Paul Van Eck, Peter Lee, Piotr Czapla, qjivy, Rholais Lii, Rodrigo Formigone, Russell Klopfer, ryantimjohn, Sang Han, SebastiáN RamíRez, shengfuintel, Siby Jose Plathottam, Silver Chan, Stanislaw Antol, Taehoon Lee, Tarang Chugh, Ted Chang, Thomas Bastiani, Xian Xu, Xiaoming (Jason) Cui, Yan Facai (颜发才), yaox12, Yashal Shakti Kanungo, Yong Tang, Yuan (Terry) Tang, Yuxin Wu, Ziyue(Louis) Lu

Release 1.7.0

Major Features And Improvements

  • Eager mode is moving out of contrib, try tf.enable_eager_execution().
  • Graph rewrites emulating fixed-point quantization compatible with TensorFlow Lite, supported by new tf.contrib.quantize package.
  • Easily customize gradient computation with tf.custom_gradient.
  • TensorBoard Debugger Plugin, the graphical user interface (GUI) of TensorFlow Debugger (tfdbg), is now in alpha.
  • Experimental support for reading a sqlite database as a Dataset with new tf.contrib.data.SqlDataset.
  • Distributed Mutex / CriticalSection added to tf.contrib.framework.CriticalSection.
  • Better text processing with tf.regex_replace